2 * Copyright (C) 2012 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com>
5 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * SPDX-License-Identifier: GPL-2.0-only
21 #include <sys/mount.h>
22 #include <sys/resource.h>
23 #include <sys/socket.h>
25 #include <sys/types.h>
27 #include <sys/resource.h>
29 #include <urcu/futex.h>
30 #include <urcu/uatomic.h>
31 #include <urcu/rculist.h>
37 #include <lttng/lttng.h>
38 #include <common/common.h>
39 #include <common/compat/poll.h>
40 #include <common/compat/socket.h>
41 #include <common/compat/endian.h>
42 #include <common/compat/getenv.h>
43 #include <common/defaults.h>
44 #include <common/daemonize.h>
45 #include <common/futex.h>
46 #include <common/sessiond-comm/sessiond-comm.h>
47 #include <common/sessiond-comm/inet.h>
48 #include <common/sessiond-comm/relayd.h>
49 #include <common/uri.h>
50 #include <common/utils.h>
51 #include <common/path.h>
52 #include <common/align.h>
53 #include <common/config/session-config.h>
54 #include <common/dynamic-buffer.h>
55 #include <common/buffer-view.h>
56 #include <common/string-utils/format.h>
57 #include <common/fd-tracker/fd-tracker.h>
58 #include <common/fd-tracker/utils.h>
60 #include "backward-compatibility-group-by.h"
62 #include "connection.h"
63 #include "ctf-trace.h"
64 #include "health-relayd.h"
67 #include "lttng-relayd.h"
69 #include "sessiond-trace-chunks.h"
71 #include "tcp_keep_alive.h"
72 #include "testpoint.h"
73 #include "tracefile-array.h"
76 #include "viewer-stream.h"
78 static const char *help_msg
=
79 #ifdef LTTNG_EMBED_HELP
80 #include <lttng-relayd.8.h>
86 enum relay_connection_status
{
87 RELAY_CONNECTION_STATUS_OK
,
88 /* An error occurred while processing an event on the connection. */
89 RELAY_CONNECTION_STATUS_ERROR
,
90 /* Connection closed/shutdown cleanly. */
91 RELAY_CONNECTION_STATUS_CLOSED
,
94 /* command line options */
95 char *opt_output_path
, *opt_working_directory
;
96 static int opt_daemon
, opt_background
, opt_print_version
, opt_allow_clear
= 1;
97 enum relay_group_output_by opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
;
100 * We need to wait for listener and live listener threads, as well as
101 * health check thread, before being ready to signal readiness.
103 #define NR_LTTNG_RELAY_READY 3
104 static int lttng_relay_ready
= NR_LTTNG_RELAY_READY
;
106 /* Size of receive buffer. */
107 #define RECV_DATA_BUFFER_SIZE 65536
109 static int recv_child_signal
; /* Set to 1 when a SIGUSR1 signal is received. */
110 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
112 static struct lttng_uri
*control_uri
;
113 static struct lttng_uri
*data_uri
;
114 static struct lttng_uri
*live_uri
;
116 const char *progname
;
118 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
119 static int tracing_group_name_override
;
121 const char * const config_section_name
= "relayd";
124 * Quit pipe for all threads. This permits a single cancellation point
125 * for all threads when receiving an event on the pipe.
127 int thread_quit_pipe
[2] = { -1, -1 };
130 * This pipe is used to inform the worker thread that a command is queued and
131 * ready to be processed.
133 static int relay_conn_pipe
[2] = { -1, -1 };
135 /* Shared between threads */
136 static int dispatch_thread_exit
;
138 static pthread_t listener_thread
;
139 static pthread_t dispatcher_thread
;
140 static pthread_t worker_thread
;
141 static pthread_t health_thread
;
144 * last_relay_stream_id_lock protects last_relay_stream_id increment
145 * atomicity on 32-bit architectures.
147 static pthread_mutex_t last_relay_stream_id_lock
= PTHREAD_MUTEX_INITIALIZER
;
148 static uint64_t last_relay_stream_id
;
151 * Relay command queue.
153 * The relay_thread_listener and relay_thread_dispatcher communicate with this
156 static struct relay_conn_queue relay_conn_queue
;
158 /* Cap of file desriptors to be in simultaneous use by the relay daemon. */
159 static unsigned int lttng_opt_fd_pool_size
= -1;
161 /* Global relay stream hash table. */
162 struct lttng_ht
*relay_streams_ht
;
164 /* Global relay viewer stream hash table. */
165 struct lttng_ht
*viewer_streams_ht
;
167 /* Global relay sessions hash table. */
168 struct lttng_ht
*sessions_ht
;
170 /* Relayd health monitoring */
171 struct health_app
*health_relayd
;
173 struct sessiond_trace_chunk_registry
*sessiond_trace_chunk_registry
;
175 /* Global fd tracker. */
176 struct fd_tracker
*the_fd_tracker
;
178 static struct option long_options
[] = {
179 { "control-port", 1, 0, 'C', },
180 { "data-port", 1, 0, 'D', },
181 { "live-port", 1, 0, 'L', },
182 { "daemonize", 0, 0, 'd', },
183 { "background", 0, 0, 'b', },
184 { "group", 1, 0, 'g', },
185 { "fd-pool-size", 1, 0, '\0', },
186 { "help", 0, 0, 'h', },
187 { "output", 1, 0, 'o', },
188 { "verbose", 0, 0, 'v', },
189 { "config", 1, 0, 'f' },
190 { "version", 0, 0, 'V' },
191 { "working-directory", 1, 0, 'w', },
192 { "group-output-by-session", 0, 0, 's', },
193 { "group-output-by-host", 0, 0, 'p', },
194 { "disallow-clear", 0, 0, 'x' },
198 static const char *config_ignore_options
[] = { "help", "config", "version" };
200 static void print_version(void) {
201 fprintf(stdout
, "%s\n", VERSION
);
204 static void relayd_config_log(void)
206 DBG("LTTng-relayd " VERSION
" - " VERSION_NAME
"%s%s",
207 GIT_VERSION
[0] == '\0' ? "" : " - " GIT_VERSION
,
208 EXTRA_VERSION_NAME
[0] == '\0' ? "" : " - " EXTRA_VERSION_NAME
);
209 if (EXTRA_VERSION_DESCRIPTION
[0] != '\0') {
210 DBG("LTTng-relayd extra version description:\n\t" EXTRA_VERSION_DESCRIPTION
"\n");
212 if (EXTRA_VERSION_PATCHES
[0] != '\0') {
213 DBG("LTTng-relayd extra patches:\n\t" EXTRA_VERSION_PATCHES
"\n");
218 * Take an option from the getopt output and set it in the right variable to be
221 * Return 0 on success else a negative value.
223 static int set_option(int opt
, const char *arg
, const char *optname
)
229 if (!strcmp(optname
, "fd-pool-size")) {
233 v
= strtoul(arg
, NULL
, 0);
234 if (errno
!= 0 || !isdigit((unsigned char) arg
[0])) {
235 ERR("Wrong value in --fd-pool-size parameter: %s", arg
);
240 ERR("File descriptor cap overflow in --fd-pool-size parameter: %s", arg
);
244 lttng_opt_fd_pool_size
= (unsigned int) v
;
246 fprintf(stderr
, "unknown option %s", optname
);
248 fprintf(stderr
, " with arg %s\n", arg
);
253 if (lttng_is_setuid_setgid()) {
254 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
255 "-C, --control-port");
257 ret
= uri_parse(arg
, &control_uri
);
259 ERR("Invalid control URI specified");
262 if (control_uri
->port
== 0) {
263 control_uri
->port
= DEFAULT_NETWORK_CONTROL_PORT
;
268 if (lttng_is_setuid_setgid()) {
269 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
272 ret
= uri_parse(arg
, &data_uri
);
274 ERR("Invalid data URI specified");
277 if (data_uri
->port
== 0) {
278 data_uri
->port
= DEFAULT_NETWORK_DATA_PORT
;
283 if (lttng_is_setuid_setgid()) {
284 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
287 ret
= uri_parse(arg
, &live_uri
);
289 ERR("Invalid live URI specified");
292 if (live_uri
->port
== 0) {
293 live_uri
->port
= DEFAULT_NETWORK_VIEWER_PORT
;
304 if (lttng_is_setuid_setgid()) {
305 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
308 tracing_group_name
= strdup(arg
);
309 if (tracing_group_name
== NULL
) {
314 tracing_group_name_override
= 1;
318 ret
= utils_show_help(8, "lttng-relayd", help_msg
);
320 ERR("Cannot show --help for `lttng-relayd`");
325 opt_print_version
= 1;
328 if (lttng_is_setuid_setgid()) {
329 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
332 ret
= asprintf(&opt_output_path
, "%s", arg
);
335 PERROR("asprintf opt_output_path");
341 if (lttng_is_setuid_setgid()) {
342 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
343 "-w, --working-directory");
345 ret
= asprintf(&opt_working_directory
, "%s", arg
);
348 PERROR("asprintf opt_working_directory");
355 /* Verbose level can increase using multiple -v */
357 lttng_opt_verbose
= config_parse_value(arg
);
359 /* Only 3 level of verbosity (-vvv). */
360 if (lttng_opt_verbose
< 3) {
361 lttng_opt_verbose
+= 1;
366 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
367 ERR("Cannot set --group-output-by-session, another --group-output-by argument is present");
370 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_SESSION
;
373 if (opt_group_output_by
!= RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
374 ERR("Cannot set --group-output-by-host, another --group-output-by argument is present");
377 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
384 /* Unknown option or other error.
385 * Error is printed by getopt, just return */
398 * config_entry_handler_cb used to handle options read from a config file.
399 * See config_entry_handler_cb comment in common/config/session-config.h for the
400 * return value conventions.
402 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
406 if (!entry
|| !entry
->name
|| !entry
->value
) {
411 /* Check if the option is to be ignored */
412 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
413 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
418 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1; i
++) {
419 /* Ignore if entry name is not fully matched. */
420 if (strcmp(entry
->name
, long_options
[i
].name
)) {
425 * If the option takes no argument on the command line,
426 * we have to check if the value is "true". We support
427 * non-zero numeric values, true, on and yes.
429 if (!long_options
[i
].has_arg
) {
430 ret
= config_parse_value(entry
->value
);
433 WARN("Invalid configuration value \"%s\" for option %s",
434 entry
->value
, entry
->name
);
436 /* False, skip boolean config option. */
441 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
445 WARN("Unrecognized option \"%s\" in daemon configuration file.",
452 static int parse_env_options(void)
457 value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_WORKING_DIRECTORY_ENV
);
459 opt_working_directory
= strdup(value
);
460 if (!opt_working_directory
) {
461 ERR("Failed to allocate working directory string (\"%s\")",
469 static int set_fd_pool_size(void)
472 struct rlimit rlimit
;
474 ret
= getrlimit(RLIMIT_NOFILE
, &rlimit
);
476 PERROR("Failed to get file descriptor limit");
481 DBG("File descriptor count limits are %" PRIu64
" (soft) and %" PRIu64
" (hard)",
482 (uint64_t) rlimit
.rlim_cur
,
483 (uint64_t) rlimit
.rlim_max
);
484 if (lttng_opt_fd_pool_size
== -1) {
485 /* Use default value (soft limit - reserve). */
486 if (rlimit
.rlim_cur
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
487 ERR("The process' file number limit is too low (%" PRIu64
"). The process' file number limit must be set to at least %i.",
488 (uint64_t) rlimit
.rlim_cur
, DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
492 lttng_opt_fd_pool_size
= rlimit
.rlim_cur
-
493 DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
497 if (lttng_opt_fd_pool_size
< DEFAULT_RELAYD_MIN_FD_POOL_SIZE
) {
498 ERR("File descriptor pool size must be set to at least %d",
499 DEFAULT_RELAYD_MIN_FD_POOL_SIZE
);
504 if (lttng_opt_fd_pool_size
> rlimit
.rlim_cur
) {
505 ERR("File descriptor pool size argument (%u) exceeds the process' soft limit (%" PRIu64
").",
506 lttng_opt_fd_pool_size
, (uint64_t) rlimit
.rlim_cur
);
511 DBG("File descriptor pool size argument (%u) adjusted to %u to accommodates transient fd uses",
512 lttng_opt_fd_pool_size
,
513 lttng_opt_fd_pool_size
- DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
);
514 lttng_opt_fd_pool_size
-= DEFAULT_RELAYD_FD_POOL_SIZE_RESERVE
;
519 static int set_options(int argc
, char **argv
)
521 int c
, ret
= 0, option_index
= 0, retval
= 0;
522 int orig_optopt
= optopt
, orig_optind
= optind
;
523 char *default_address
, *optstring
;
524 const char *config_path
= NULL
;
526 optstring
= utils_generate_optstring(long_options
,
527 sizeof(long_options
) / sizeof(struct option
));
533 /* Check for the --config option */
535 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
536 &option_index
)) != -1) {
540 } else if (c
!= 'f') {
544 if (lttng_is_setuid_setgid()) {
545 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
548 config_path
= utils_expand_path(optarg
);
550 ERR("Failed to resolve path: %s", optarg
);
555 ret
= config_get_section_entries(config_path
, config_section_name
,
556 config_entry_handler
, NULL
);
559 ERR("Invalid configuration option at line %i", ret
);
565 /* Reset getopt's global state */
566 optopt
= orig_optopt
;
567 optind
= orig_optind
;
569 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
574 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
581 /* assign default values */
582 if (control_uri
== NULL
) {
583 ret
= asprintf(&default_address
,
584 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS
":%d",
585 DEFAULT_NETWORK_CONTROL_PORT
);
587 PERROR("asprintf default data address");
592 ret
= uri_parse(default_address
, &control_uri
);
593 free(default_address
);
595 ERR("Invalid control URI specified");
600 if (data_uri
== NULL
) {
601 ret
= asprintf(&default_address
,
602 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS
":%d",
603 DEFAULT_NETWORK_DATA_PORT
);
605 PERROR("asprintf default data address");
610 ret
= uri_parse(default_address
, &data_uri
);
611 free(default_address
);
613 ERR("Invalid data URI specified");
618 if (live_uri
== NULL
) {
619 ret
= asprintf(&default_address
,
620 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
":%d",
621 DEFAULT_NETWORK_VIEWER_PORT
);
623 PERROR("asprintf default viewer control address");
628 ret
= uri_parse(default_address
, &live_uri
);
629 free(default_address
);
631 ERR("Invalid viewer control URI specified");
636 ret
= set_fd_pool_size();
642 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_UNKNOWN
) {
643 opt_group_output_by
= RELAYD_GROUP_OUTPUT_BY_HOST
;
645 if (opt_allow_clear
) {
646 /* Check if env variable exists. */
647 const char *value
= lttng_secure_getenv(DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
649 ret
= config_parse_value(value
);
651 ERR("Invalid value for %s specified", DEFAULT_LTTNG_RELAYD_DISALLOW_CLEAR_ENV
);
655 opt_allow_clear
= !ret
;
664 static void print_global_objects(void)
666 print_viewer_streams();
667 print_relay_streams();
671 static int noop_close(void *data
, int *fds
)
676 static void untrack_stdio(void)
678 int fds
[] = { fileno(stdout
), fileno(stderr
) };
681 * noop_close is used since we don't really want to close
682 * the stdio output fds; we merely want to stop tracking them.
684 (void) fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
685 fds
, 2, noop_close
, NULL
);
691 static void relayd_cleanup(void)
693 print_global_objects();
697 if (viewer_streams_ht
)
698 lttng_ht_destroy(viewer_streams_ht
);
699 if (relay_streams_ht
)
700 lttng_ht_destroy(relay_streams_ht
);
702 lttng_ht_destroy(sessions_ht
);
704 free(opt_output_path
);
705 free(opt_working_directory
);
708 health_app_destroy(health_relayd
);
710 /* Close thread quit pipes */
711 if (health_quit_pipe
[0] != -1) {
712 (void) fd_tracker_util_pipe_close(
713 the_fd_tracker
, health_quit_pipe
);
715 if (thread_quit_pipe
[0] != -1) {
716 (void) fd_tracker_util_pipe_close(
717 the_fd_tracker
, thread_quit_pipe
);
719 if (sessiond_trace_chunk_registry
) {
720 sessiond_trace_chunk_registry_destroy(
721 sessiond_trace_chunk_registry
);
723 if (the_fd_tracker
) {
726 * fd_tracker_destroy() will log the contents of the fd-tracker
727 * if a leak is detected.
729 fd_tracker_destroy(the_fd_tracker
);
732 uri_free(control_uri
);
734 /* Live URI is freed in the live thread. */
736 if (tracing_group_name_override
) {
737 free((void *) tracing_group_name
);
742 * Write to writable pipe used to notify a thread.
744 static int notify_thread_pipe(int wpipe
)
748 ret
= lttng_write(wpipe
, "!", 1);
750 PERROR("write poll pipe");
758 static int notify_health_quit_pipe(int *pipe
)
762 ret
= lttng_write(pipe
[1], "4", 1);
764 PERROR("write relay health quit");
773 * Stop all relayd and relayd-live threads.
775 int lttng_relay_stop_threads(void)
779 /* Stopping all threads */
780 DBG("Terminating all threads");
781 if (notify_thread_pipe(thread_quit_pipe
[1])) {
782 ERR("write error on thread quit pipe");
786 if (notify_health_quit_pipe(health_quit_pipe
)) {
787 ERR("write error on health quit pipe");
790 /* Dispatch thread */
791 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
792 futex_nto1_wake(&relay_conn_queue
.futex
);
794 if (relayd_live_stop()) {
795 ERR("Error stopping live threads");
802 * Signal handler for the daemon
804 * Simply stop all worker threads, leaving main() return gracefully after
805 * joining all threads and calling cleanup().
807 static void sighandler(int sig
)
811 DBG("SIGINT caught");
812 if (lttng_relay_stop_threads()) {
813 ERR("Error stopping threads");
817 DBG("SIGTERM caught");
818 if (lttng_relay_stop_threads()) {
819 ERR("Error stopping threads");
823 CMM_STORE_SHARED(recv_child_signal
, 1);
831 * Setup signal handler for :
832 * SIGINT, SIGTERM, SIGPIPE
834 static int set_signal_handler(void)
840 if ((ret
= sigemptyset(&sigset
)) < 0) {
841 PERROR("sigemptyset");
848 sa
.sa_handler
= sighandler
;
849 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
854 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
859 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
864 sa
.sa_handler
= SIG_IGN
;
865 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
870 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
875 void lttng_relay_notify_ready(void)
877 /* Notify the parent of the fork() process that we are ready. */
878 if (opt_daemon
|| opt_background
) {
879 if (uatomic_sub_return(<tng_relay_ready
, 1) == 0) {
880 kill(child_ppid
, SIGUSR1
);
886 * Init thread quit pipe.
888 * Return -1 on error or 0 if all pipes are created.
890 static int init_thread_quit_pipe(void)
892 return fd_tracker_util_pipe_open_cloexec(
893 the_fd_tracker
, "Quit pipe", thread_quit_pipe
);
897 * Init health quit pipe.
899 * Return -1 on error or 0 if all pipes are created.
901 static int init_health_quit_pipe(void)
903 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
904 "Health quit pipe", health_quit_pipe
);
908 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
910 static int create_named_thread_poll_set(struct lttng_poll_event
*events
,
911 int size
, const char *name
)
915 if (events
== NULL
|| size
== 0) {
920 ret
= fd_tracker_util_poll_create(the_fd_tracker
,
921 name
, events
, 1, LTTNG_CLOEXEC
);
923 PERROR("Failed to create \"%s\" poll file descriptor", name
);
928 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
940 * Check if the thread quit pipe was triggered.
942 * Return 1 if it was triggered else 0;
944 static int check_thread_quit_pipe(int fd
, uint32_t events
)
946 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
953 static int create_sock(void *data
, int *out_fd
)
956 struct lttcomm_sock
*sock
= data
;
958 ret
= lttcomm_create_sock(sock
);
968 static int close_sock(void *data
, int *in_fd
)
970 struct lttcomm_sock
*sock
= data
;
972 return sock
->ops
->close(sock
);
975 static int accept_sock(void *data
, int *out_fd
)
978 /* Socks is an array of in_sock, out_sock. */
979 struct lttcomm_sock
**socks
= data
;
980 struct lttcomm_sock
*in_sock
= socks
[0];
982 socks
[1] = in_sock
->ops
->accept(in_sock
);
987 *out_fd
= socks
[1]->fd
;
993 * Create and init socket from uri.
995 static struct lttcomm_sock
*relay_socket_create(struct lttng_uri
*uri
,
999 struct lttcomm_sock
*sock
= NULL
;
1000 char uri_str
[PATH_MAX
];
1001 char *formated_name
= NULL
;
1003 sock
= lttcomm_alloc_sock_from_uri(uri
);
1005 ERR("Allocating socket");
1010 * Don't fail to create the socket if the name can't be built as it is
1011 * only used for debugging purposes.
1013 ret
= uri_to_str_url(uri
, uri_str
, sizeof(uri_str
));
1014 uri_str
[sizeof(uri_str
) - 1] = '\0';
1016 ret
= asprintf(&formated_name
, "%s socket @ %s", name
,
1019 formated_name
= NULL
;
1023 ret
= fd_tracker_open_unsuspendable_fd(the_fd_tracker
, &sock_fd
,
1024 (const char **) (formated_name
? &formated_name
: NULL
),
1025 1, create_sock
, sock
);
1027 PERROR("Failed to open \"%s\" relay socket",
1028 formated_name
?: "Unknown");
1031 DBG("Listening on %s socket %d", name
, sock
->fd
);
1033 ret
= sock
->ops
->bind(sock
);
1035 PERROR("Failed to bind socket");
1039 ret
= sock
->ops
->listen(sock
, -1);
1045 free(formated_name
);
1050 lttcomm_destroy_sock(sock
);
1052 free(formated_name
);
1057 struct lttcomm_sock
*accept_relayd_sock(struct lttcomm_sock
*listening_sock
,
1061 struct lttcomm_sock
*socks
[2] = { listening_sock
, NULL
};
1062 struct lttcomm_sock
*new_sock
= NULL
;
1064 ret
= fd_tracker_open_unsuspendable_fd(
1065 the_fd_tracker
, &out_fd
,
1066 (const char **) &name
,
1067 1, accept_sock
, &socks
);
1071 new_sock
= socks
[1];
1072 DBG("%s accepted, socket %d", name
, new_sock
->fd
);
1078 * This thread manages the listening for new connections on the network
1080 static void *relay_thread_listener(void *data
)
1082 int i
, ret
, pollfd
, err
= -1;
1083 uint32_t revents
, nb_fd
;
1084 struct lttng_poll_event events
;
1085 struct lttcomm_sock
*control_sock
, *data_sock
;
1087 DBG("[thread] Relay listener started");
1089 rcu_register_thread();
1090 health_register(health_relayd
, HEALTH_RELAYD_TYPE_LISTENER
);
1092 health_code_update();
1094 control_sock
= relay_socket_create(control_uri
, "Control listener");
1095 if (!control_sock
) {
1096 goto error_sock_control
;
1099 data_sock
= relay_socket_create(data_uri
, "Data listener");
1101 goto error_sock_relay
;
1105 * Pass 3 as size here for the thread quit pipe, control and
1108 ret
= create_named_thread_poll_set(&events
, 3, "Listener thread epoll");
1110 goto error_create_poll
;
1113 /* Add the control socket */
1114 ret
= lttng_poll_add(&events
, control_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1116 goto error_poll_add
;
1119 /* Add the data socket */
1120 ret
= lttng_poll_add(&events
, data_sock
->fd
, LPOLLIN
| LPOLLRDHUP
);
1122 goto error_poll_add
;
1125 lttng_relay_notify_ready();
1127 if (testpoint(relayd_thread_listener
)) {
1128 goto error_testpoint
;
1132 health_code_update();
1134 DBG("Listener accepting connections");
1137 health_poll_entry();
1138 ret
= lttng_poll_wait(&events
, -1);
1142 * Restart interrupted system call.
1144 if (errno
== EINTR
) {
1152 DBG("Relay new connection received");
1153 for (i
= 0; i
< nb_fd
; i
++) {
1154 health_code_update();
1156 /* Fetch once the poll data */
1157 revents
= LTTNG_POLL_GETEV(&events
, i
);
1158 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1160 /* Thread quit pipe has been closed. Killing thread. */
1161 ret
= check_thread_quit_pipe(pollfd
, revents
);
1167 if (revents
& LPOLLIN
) {
1169 * A new connection is requested, therefore a
1170 * sessiond/consumerd connection is allocated in
1171 * this thread, enqueued to a global queue and
1172 * dequeued (and freed) in the worker thread.
1175 struct relay_connection
*new_conn
;
1176 struct lttcomm_sock
*newsock
= NULL
;
1177 enum connection_type type
;
1179 if (pollfd
== data_sock
->fd
) {
1181 newsock
= accept_relayd_sock(data_sock
,
1182 "Data socket to relayd");
1184 assert(pollfd
== control_sock
->fd
);
1185 type
= RELAY_CONTROL
;
1186 newsock
= accept_relayd_sock(control_sock
,
1187 "Control socket to relayd");
1190 PERROR("accepting sock");
1194 ret
= setsockopt(newsock
->fd
, SOL_SOCKET
, SO_REUSEADDR
, &val
,
1197 PERROR("setsockopt inet");
1198 lttcomm_destroy_sock(newsock
);
1202 ret
= socket_apply_keep_alive_config(newsock
->fd
);
1204 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
1206 lttcomm_destroy_sock(newsock
);
1210 new_conn
= connection_create(newsock
, type
);
1212 lttcomm_destroy_sock(newsock
);
1216 /* Enqueue request for the dispatcher thread. */
1217 cds_wfcq_enqueue(&relay_conn_queue
.head
, &relay_conn_queue
.tail
,
1221 * Wake the dispatch queue futex.
1222 * Implicit memory barrier with the
1223 * exchange in cds_wfcq_enqueue.
1225 futex_nto1_wake(&relay_conn_queue
.futex
);
1226 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1227 ERR("socket poll error");
1230 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1240 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
1242 if (data_sock
->fd
>= 0) {
1243 int data_sock_fd
= data_sock
->fd
;
1245 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1246 &data_sock_fd
, 1, close_sock
,
1249 PERROR("Failed to close the data listener socket file descriptor");
1253 lttcomm_destroy_sock(data_sock
);
1255 if (control_sock
->fd
>= 0) {
1256 int control_sock_fd
= control_sock
->fd
;
1258 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
,
1259 &control_sock_fd
, 1, close_sock
,
1262 PERROR("Failed to close the control listener socket file descriptor");
1264 control_sock
->fd
= -1;
1266 lttcomm_destroy_sock(control_sock
);
1270 ERR("Health error occurred in %s", __func__
);
1272 health_unregister(health_relayd
);
1273 rcu_unregister_thread();
1274 DBG("Relay listener thread cleanup complete");
1275 lttng_relay_stop_threads();
1280 * This thread manages the dispatching of the requests to worker threads
1282 static void *relay_thread_dispatcher(void *data
)
1286 struct cds_wfcq_node
*node
;
1287 struct relay_connection
*new_conn
= NULL
;
1289 DBG("[thread] Relay dispatcher started");
1291 health_register(health_relayd
, HEALTH_RELAYD_TYPE_DISPATCHER
);
1293 if (testpoint(relayd_thread_dispatcher
)) {
1294 goto error_testpoint
;
1297 health_code_update();
1300 health_code_update();
1302 /* Atomically prepare the queue futex */
1303 futex_nto1_prepare(&relay_conn_queue
.futex
);
1305 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1310 health_code_update();
1312 /* Dequeue commands */
1313 node
= cds_wfcq_dequeue_blocking(&relay_conn_queue
.head
,
1314 &relay_conn_queue
.tail
);
1316 DBG("Woken up but nothing in the relay command queue");
1317 /* Continue thread execution */
1320 new_conn
= caa_container_of(node
, struct relay_connection
, qnode
);
1322 DBG("Dispatching request waiting on sock %d", new_conn
->sock
->fd
);
1325 * Inform worker thread of the new request. This
1326 * call is blocking so we can be assured that
1327 * the data will be read at some point in time
1328 * or wait to the end of the world :)
1330 ret
= lttng_write(relay_conn_pipe
[1], &new_conn
, sizeof(new_conn
));
1332 PERROR("write connection pipe");
1333 connection_put(new_conn
);
1336 } while (node
!= NULL
);
1338 /* Futex wait on queue. Blocking call on futex() */
1339 health_poll_entry();
1340 futex_nto1_wait(&relay_conn_queue
.futex
);
1344 /* Normal exit, no error */
1351 ERR("Health error occurred in %s", __func__
);
1353 health_unregister(health_relayd
);
1354 DBG("Dispatch thread dying");
1355 lttng_relay_stop_threads();
1359 static bool session_streams_have_index(const struct relay_session
*session
)
1361 return session
->minor
>= 4 && !session
->snapshot
;
1365 * Handle the RELAYD_CREATE_SESSION command.
1367 * On success, send back the session id or else return a negative value.
1369 static int relay_create_session(const struct lttcomm_relayd_hdr
*recv_hdr
,
1370 struct relay_connection
*conn
,
1371 const struct lttng_buffer_view
*payload
)
1375 struct relay_session
*session
= NULL
;
1376 struct lttcomm_relayd_create_session_reply_2_11 reply
= {};
1377 char session_name
[LTTNG_NAME_MAX
] = {};
1378 char hostname
[LTTNG_HOST_NAME_MAX
] = {};
1379 uint32_t live_timer
= 0;
1380 bool snapshot
= false;
1381 bool session_name_contains_creation_timestamp
= false;
1382 /* Left nil for peers < 2.11. */
1383 char base_path
[LTTNG_PATH_MAX
] = {};
1384 lttng_uuid sessiond_uuid
= {};
1385 LTTNG_OPTIONAL(uint64_t) id_sessiond
= {};
1386 LTTNG_OPTIONAL(uint64_t) current_chunk_id
= {};
1387 LTTNG_OPTIONAL(time_t) creation_time
= {};
1388 struct lttng_dynamic_buffer reply_payload
;
1390 lttng_dynamic_buffer_init(&reply_payload
);
1392 if (conn
->minor
< 4) {
1393 /* From 2.1 to 2.3 */
1395 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1396 /* From 2.4 to 2.10 */
1397 ret
= cmd_create_session_2_4(payload
, session_name
,
1398 hostname
, &live_timer
, &snapshot
);
1400 bool has_current_chunk
;
1401 uint64_t current_chunk_id_value
;
1402 time_t creation_time_value
;
1403 uint64_t id_sessiond_value
;
1405 /* From 2.11 to ... */
1406 ret
= cmd_create_session_2_11(payload
, session_name
, hostname
,
1407 base_path
, &live_timer
, &snapshot
, &id_sessiond_value
,
1408 sessiond_uuid
, &has_current_chunk
,
1409 ¤t_chunk_id_value
, &creation_time_value
,
1410 &session_name_contains_creation_timestamp
);
1411 if (lttng_uuid_is_nil(sessiond_uuid
)) {
1412 /* The nil UUID is reserved for pre-2.11 clients. */
1413 ERR("Illegal nil UUID announced by peer in create session command");
1417 LTTNG_OPTIONAL_SET(&id_sessiond
, id_sessiond_value
);
1418 LTTNG_OPTIONAL_SET(&creation_time
, creation_time_value
);
1419 if (has_current_chunk
) {
1420 LTTNG_OPTIONAL_SET(¤t_chunk_id
,
1421 current_chunk_id_value
);
1429 session
= session_create(session_name
, hostname
, base_path
, live_timer
,
1430 snapshot
, sessiond_uuid
,
1431 id_sessiond
.is_set
? &id_sessiond
.value
: NULL
,
1432 current_chunk_id
.is_set
? ¤t_chunk_id
.value
: NULL
,
1433 creation_time
.is_set
? &creation_time
.value
: NULL
,
1434 conn
->major
, conn
->minor
,
1435 session_name_contains_creation_timestamp
);
1440 assert(!conn
->session
);
1441 conn
->session
= session
;
1442 DBG("Created session %" PRIu64
, session
->id
);
1444 reply
.generic
.session_id
= htobe64(session
->id
);
1448 reply
.generic
.ret_code
= htobe32(LTTNG_ERR_FATAL
);
1450 reply
.generic
.ret_code
= htobe32(LTTNG_OK
);
1453 if (conn
->minor
< 11) {
1454 /* From 2.1 to 2.10 */
1455 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1456 &reply
.generic
, sizeof(reply
.generic
));
1458 ERR("Failed to append \"create session\" command reply header to payload buffer");
1463 const uint32_t output_path_length
=
1464 session
? strlen(session
->output_path
) + 1 : 0;
1466 reply
.output_path_length
= htobe32(output_path_length
);
1467 ret
= lttng_dynamic_buffer_append(
1468 &reply_payload
, &reply
, sizeof(reply
));
1470 ERR("Failed to append \"create session\" command reply header to payload buffer");
1474 if (output_path_length
) {
1475 ret
= lttng_dynamic_buffer_append(&reply_payload
,
1476 session
->output_path
,
1477 output_path_length
);
1479 ERR("Failed to append \"create session\" command reply path to payload buffer");
1485 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, reply_payload
.data
,
1486 reply_payload
.size
, 0);
1487 if (send_ret
< (ssize_t
) reply_payload
.size
) {
1488 ERR("Failed to send \"create session\" command reply of %zu bytes (ret = %zd)",
1489 reply_payload
.size
, send_ret
);
1493 if (ret
< 0 && session
) {
1494 session_put(session
);
1496 lttng_dynamic_buffer_reset(&reply_payload
);
1501 * When we have received all the streams and the metadata for a channel,
1502 * we make them visible to the viewer threads.
1504 static void publish_connection_local_streams(struct relay_connection
*conn
)
1506 struct relay_stream
*stream
;
1507 struct relay_session
*session
= conn
->session
;
1510 * We publish all streams belonging to a session atomically wrt
1513 pthread_mutex_lock(&session
->lock
);
1515 cds_list_for_each_entry_rcu(stream
, &session
->recv_list
,
1517 stream_publish(stream
);
1522 * Inform the viewer that there are new streams in the session.
1524 if (session
->viewer_attached
) {
1525 uatomic_set(&session
->new_streams
, 1);
1528 pthread_mutex_unlock(&session
->lock
);
1531 static int conform_channel_path(char *channel_path
)
1535 if (strstr("../", channel_path
)) {
1536 ERR("Refusing channel path as it walks up the path hierarchy: \"%s\"",
1542 if (*channel_path
== '/') {
1543 const size_t len
= strlen(channel_path
);
1546 * Channel paths from peers prior to 2.11 are expressed as an
1547 * absolute path that is, in reality, relative to the relay
1548 * daemon's output directory. Remove the leading slash so it
1549 * is correctly interpreted as a relative path later on.
1551 * len (and not len - 1) is used to copy the trailing NULL.
1553 bcopy(channel_path
+ 1, channel_path
, len
);
1560 * relay_add_stream: allocate a new stream for a session
1562 static int relay_add_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1563 struct relay_connection
*conn
,
1564 const struct lttng_buffer_view
*payload
)
1568 struct relay_session
*session
= conn
->session
;
1569 struct relay_stream
*stream
= NULL
;
1570 struct lttcomm_relayd_status_stream reply
;
1571 struct ctf_trace
*trace
= NULL
;
1572 uint64_t stream_handle
= -1ULL;
1573 char *path_name
= NULL
, *channel_name
= NULL
;
1574 uint64_t tracefile_size
= 0, tracefile_count
= 0;
1575 LTTNG_OPTIONAL(uint64_t) stream_chunk_id
= {};
1577 if (!session
|| !conn
->version_check_done
) {
1578 ERR("Trying to add a stream before version check");
1580 goto end_no_session
;
1583 if (session
->minor
== 1) {
1585 ret
= cmd_recv_stream_2_1(payload
, &path_name
,
1587 } else if (session
->minor
> 1 && session
->minor
< 11) {
1588 /* From 2.2 to 2.10 */
1589 ret
= cmd_recv_stream_2_2(payload
, &path_name
,
1590 &channel_name
, &tracefile_size
, &tracefile_count
);
1592 /* From 2.11 to ... */
1593 ret
= cmd_recv_stream_2_11(payload
, &path_name
,
1594 &channel_name
, &tracefile_size
, &tracefile_count
,
1595 &stream_chunk_id
.value
);
1596 stream_chunk_id
.is_set
= true;
1603 if (conform_channel_path(path_name
)) {
1608 * Backward compatibility for --group-output-by-session.
1609 * Prior to lttng 2.11, the complete path is passed by the stream.
1610 * Starting at 2.11, lttng-relayd uses chunk. When dealing with producer
1611 * >=2.11 the chunk is responsible for the output path. When dealing
1612 * with producer < 2.11 the chunk output_path is the root output path
1613 * and the stream carries the complete path (path_name).
1614 * To support --group-output-by-session with older producer (<2.11), we
1615 * need to craft the path based on the stream path.
1617 if (opt_group_output_by
== RELAYD_GROUP_OUTPUT_BY_SESSION
) {
1618 if (conn
->minor
< 4) {
1620 * From 2.1 to 2.3, the session_name is not passed on
1621 * the RELAYD_CREATE_SESSION command. The session name
1622 * is necessary to detect the presence of a base_path
1623 * inside the stream path. Without it we cannot perform
1624 * a valid group-output-by-session transformation.
1626 WARN("Unable to perform a --group-by-session transformation for session %" PRIu64
1627 " for stream with path \"%s\" as it is produced by a peer using a protocol older than v2.4",
1628 session
->id
, path_name
);
1629 } else if (conn
->minor
>= 4 && conn
->minor
< 11) {
1630 char *group_by_session_path_name
;
1632 assert(session
->session_name
[0] != '\0');
1634 group_by_session_path_name
=
1635 backward_compat_group_by_session(
1637 session
->session_name
,
1638 session
->creation_time
.value
);
1639 if (!group_by_session_path_name
) {
1640 ERR("Failed to apply group by session to stream of session %" PRIu64
,
1645 DBG("Transformed session path from \"%s\" to \"%s\" to honor per-session name grouping",
1646 path_name
, group_by_session_path_name
);
1649 path_name
= group_by_session_path_name
;
1653 trace
= ctf_trace_get_by_path_or_create(session
, path_name
);
1658 /* This stream here has one reference on the trace. */
1659 pthread_mutex_lock(&last_relay_stream_id_lock
);
1660 stream_handle
= ++last_relay_stream_id
;
1661 pthread_mutex_unlock(&last_relay_stream_id_lock
);
1663 /* We pass ownership of path_name and channel_name. */
1664 stream
= stream_create(trace
, stream_handle
, path_name
,
1665 channel_name
, tracefile_size
, tracefile_count
);
1667 channel_name
= NULL
;
1670 * Streams are the owners of their trace. Reference to trace is
1671 * kept within stream_create().
1673 ctf_trace_put(trace
);
1676 memset(&reply
, 0, sizeof(reply
));
1677 reply
.handle
= htobe64(stream_handle
);
1679 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1681 reply
.ret_code
= htobe32(LTTNG_OK
);
1684 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1685 sizeof(struct lttcomm_relayd_status_stream
), 0);
1686 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1687 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1699 * relay_close_stream: close a specific stream
1701 static int relay_close_stream(const struct lttcomm_relayd_hdr
*recv_hdr
,
1702 struct relay_connection
*conn
,
1703 const struct lttng_buffer_view
*payload
)
1707 struct relay_session
*session
= conn
->session
;
1708 struct lttcomm_relayd_close_stream stream_info
;
1709 struct lttcomm_relayd_generic_reply reply
;
1710 struct relay_stream
*stream
;
1712 DBG("Close stream received");
1714 if (!session
|| !conn
->version_check_done
) {
1715 ERR("Trying to close a stream before version check");
1717 goto end_no_session
;
1720 if (payload
->size
< sizeof(stream_info
)) {
1721 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1722 sizeof(stream_info
), payload
->size
);
1724 goto end_no_session
;
1726 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1727 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1728 stream_info
.last_net_seq_num
= be64toh(stream_info
.last_net_seq_num
);
1730 stream
= stream_get_by_id(stream_info
.stream_id
);
1737 * Set last_net_seq_num before the close flag. Required by data
1740 pthread_mutex_lock(&stream
->lock
);
1741 stream
->last_net_seq_num
= stream_info
.last_net_seq_num
;
1742 pthread_mutex_unlock(&stream
->lock
);
1745 * This is one of the conditions which may trigger a stream close
1746 * with the others being:
1747 * 1) A close command is received for a stream
1748 * 2) The control connection owning the stream is closed
1749 * 3) We have received all of the stream's data _after_ a close
1752 try_stream_close(stream
);
1757 memset(&reply
, 0, sizeof(reply
));
1759 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1761 reply
.ret_code
= htobe32(LTTNG_OK
);
1763 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1764 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1765 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1766 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1776 * relay_reset_metadata: reset a metadata stream
1779 int relay_reset_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1780 struct relay_connection
*conn
,
1781 const struct lttng_buffer_view
*payload
)
1785 struct relay_session
*session
= conn
->session
;
1786 struct lttcomm_relayd_reset_metadata stream_info
;
1787 struct lttcomm_relayd_generic_reply reply
;
1788 struct relay_stream
*stream
;
1790 DBG("Reset metadata received");
1792 if (!session
|| !conn
->version_check_done
) {
1793 ERR("Trying to reset a metadata stream before version check");
1795 goto end_no_session
;
1798 if (payload
->size
< sizeof(stream_info
)) {
1799 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1800 sizeof(stream_info
), payload
->size
);
1802 goto end_no_session
;
1804 memcpy(&stream_info
, payload
->data
, sizeof(stream_info
));
1805 stream_info
.stream_id
= be64toh(stream_info
.stream_id
);
1806 stream_info
.version
= be64toh(stream_info
.version
);
1808 DBG("Update metadata to version %" PRIu64
, stream_info
.version
);
1810 /* Unsupported for live sessions for now. */
1811 if (session
->live_timer
!= 0) {
1816 stream
= stream_get_by_id(stream_info
.stream_id
);
1821 pthread_mutex_lock(&stream
->lock
);
1822 if (!stream
->is_metadata
) {
1827 ret
= stream_reset_file(stream
);
1829 ERR("Failed to reset metadata stream %" PRIu64
1830 ": stream_path = %s, channel = %s",
1831 stream
->stream_handle
, stream
->path_name
,
1832 stream
->channel_name
);
1836 pthread_mutex_unlock(&stream
->lock
);
1840 memset(&reply
, 0, sizeof(reply
));
1842 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1844 reply
.ret_code
= htobe32(LTTNG_OK
);
1846 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1847 sizeof(struct lttcomm_relayd_generic_reply
), 0);
1848 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1849 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1859 * relay_unknown_command: send -1 if received unknown command
1861 static void relay_unknown_command(struct relay_connection
*conn
)
1863 struct lttcomm_relayd_generic_reply reply
;
1866 memset(&reply
, 0, sizeof(reply
));
1867 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
1868 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
1869 if (send_ret
< sizeof(reply
)) {
1870 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret
);
1875 * relay_start: send an acknowledgment to the client to tell if we are
1876 * ready to receive data. We are ready if a session is established.
1878 static int relay_start(const struct lttcomm_relayd_hdr
*recv_hdr
,
1879 struct relay_connection
*conn
,
1880 const struct lttng_buffer_view
*payload
)
1884 struct lttcomm_relayd_generic_reply reply
;
1885 struct relay_session
*session
= conn
->session
;
1888 DBG("Trying to start the streaming without a session established");
1889 ret
= htobe32(LTTNG_ERR_UNK
);
1892 memset(&reply
, 0, sizeof(reply
));
1893 reply
.ret_code
= htobe32(LTTNG_OK
);
1894 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
1896 if (send_ret
< (ssize_t
) sizeof(reply
)) {
1897 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1906 * relay_recv_metadata: receive the metadata for the session.
1908 static int relay_recv_metadata(const struct lttcomm_relayd_hdr
*recv_hdr
,
1909 struct relay_connection
*conn
,
1910 const struct lttng_buffer_view
*payload
)
1913 struct relay_session
*session
= conn
->session
;
1914 struct lttcomm_relayd_metadata_payload metadata_payload_header
;
1915 struct relay_stream
*metadata_stream
;
1916 uint64_t metadata_payload_size
;
1917 struct lttng_buffer_view packet_view
;
1920 ERR("Metadata sent before version check");
1925 if (recv_hdr
->data_size
< sizeof(struct lttcomm_relayd_metadata_payload
)) {
1926 ERR("Incorrect data size");
1930 metadata_payload_size
= recv_hdr
->data_size
-
1931 sizeof(struct lttcomm_relayd_metadata_payload
);
1933 memcpy(&metadata_payload_header
, payload
->data
,
1934 sizeof(metadata_payload_header
));
1935 metadata_payload_header
.stream_id
= be64toh(
1936 metadata_payload_header
.stream_id
);
1937 metadata_payload_header
.padding_size
= be32toh(
1938 metadata_payload_header
.padding_size
);
1940 metadata_stream
= stream_get_by_id(metadata_payload_header
.stream_id
);
1941 if (!metadata_stream
) {
1946 packet_view
= lttng_buffer_view_from_view(payload
,
1947 sizeof(metadata_payload_header
), metadata_payload_size
);
1948 if (!packet_view
.data
) {
1949 ERR("Invalid metadata packet length announced by header");
1954 pthread_mutex_lock(&metadata_stream
->lock
);
1955 ret
= stream_write(metadata_stream
, &packet_view
,
1956 metadata_payload_header
.padding_size
);
1957 pthread_mutex_unlock(&metadata_stream
->lock
);
1963 stream_put(metadata_stream
);
1969 * relay_send_version: send relayd version number
1971 static int relay_send_version(const struct lttcomm_relayd_hdr
*recv_hdr
,
1972 struct relay_connection
*conn
,
1973 const struct lttng_buffer_view
*payload
)
1977 struct lttcomm_relayd_version reply
, msg
;
1978 bool compatible
= true;
1980 conn
->version_check_done
= true;
1982 /* Get version from the other side. */
1983 if (payload
->size
< sizeof(msg
)) {
1984 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1985 sizeof(msg
), payload
->size
);
1990 memcpy(&msg
, payload
->data
, sizeof(msg
));
1991 msg
.major
= be32toh(msg
.major
);
1992 msg
.minor
= be32toh(msg
.minor
);
1994 memset(&reply
, 0, sizeof(reply
));
1995 reply
.major
= RELAYD_VERSION_COMM_MAJOR
;
1996 reply
.minor
= RELAYD_VERSION_COMM_MINOR
;
1998 /* Major versions must be the same */
1999 if (reply
.major
!= msg
.major
) {
2000 DBG("Incompatible major versions (%u vs %u), deleting session",
2001 reply
.major
, msg
.major
);
2005 conn
->major
= reply
.major
;
2006 /* We adapt to the lowest compatible version */
2007 if (reply
.minor
<= msg
.minor
) {
2008 conn
->minor
= reply
.minor
;
2010 conn
->minor
= msg
.minor
;
2013 reply
.major
= htobe32(reply
.major
);
2014 reply
.minor
= htobe32(reply
.minor
);
2015 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2017 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2018 ERR("Failed to send \"send version\" command reply (ret = %zd)",
2031 DBG("Version check done using protocol %u.%u", conn
->major
,
2039 * Check for data pending for a given stream id from the session daemon.
2041 static int relay_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2042 struct relay_connection
*conn
,
2043 const struct lttng_buffer_view
*payload
)
2045 struct relay_session
*session
= conn
->session
;
2046 struct lttcomm_relayd_data_pending msg
;
2047 struct lttcomm_relayd_generic_reply reply
;
2048 struct relay_stream
*stream
;
2051 uint64_t stream_seq
;
2053 DBG("Data pending command received");
2055 if (!session
|| !conn
->version_check_done
) {
2056 ERR("Trying to check for data before version check");
2058 goto end_no_session
;
2061 if (payload
->size
< sizeof(msg
)) {
2062 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
2063 sizeof(msg
), payload
->size
);
2065 goto end_no_session
;
2067 memcpy(&msg
, payload
->data
, sizeof(msg
));
2068 msg
.stream_id
= be64toh(msg
.stream_id
);
2069 msg
.last_net_seq_num
= be64toh(msg
.last_net_seq_num
);
2071 stream
= stream_get_by_id(msg
.stream_id
);
2072 if (stream
== NULL
) {
2077 pthread_mutex_lock(&stream
->lock
);
2079 if (session_streams_have_index(session
)) {
2081 * Ensure that both the index and stream data have been
2082 * flushed up to the requested point.
2084 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2086 stream_seq
= stream
->prev_data_seq
;
2088 DBG("Data pending for stream id %" PRIu64
": prev_data_seq %" PRIu64
2089 ", prev_index_seq %" PRIu64
2090 ", and last_seq %" PRIu64
, msg
.stream_id
,
2091 stream
->prev_data_seq
, stream
->prev_index_seq
,
2092 msg
.last_net_seq_num
);
2094 /* Avoid wrapping issue */
2095 if (((int64_t) (stream_seq
- msg
.last_net_seq_num
)) >= 0) {
2096 /* Data has in fact been written and is NOT pending */
2099 /* Data still being streamed thus pending */
2103 stream
->data_pending_check_done
= true;
2104 pthread_mutex_unlock(&stream
->lock
);
2109 memset(&reply
, 0, sizeof(reply
));
2110 reply
.ret_code
= htobe32(ret
);
2111 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2112 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2113 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
2123 * Wait for the control socket to reach a quiescent state.
2125 * Note that for now, when receiving this command from the session
2126 * daemon, this means that every subsequent commands or data received on
2127 * the control socket has been handled. So, this is why we simply return
2130 static int relay_quiescent_control(const struct lttcomm_relayd_hdr
*recv_hdr
,
2131 struct relay_connection
*conn
,
2132 const struct lttng_buffer_view
*payload
)
2136 struct relay_stream
*stream
;
2137 struct lttcomm_relayd_quiescent_control msg
;
2138 struct lttcomm_relayd_generic_reply reply
;
2140 DBG("Checking quiescent state on control socket");
2142 if (!conn
->session
|| !conn
->version_check_done
) {
2143 ERR("Trying to check for data before version check");
2145 goto end_no_session
;
2148 if (payload
->size
< sizeof(msg
)) {
2149 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2150 sizeof(msg
), payload
->size
);
2152 goto end_no_session
;
2154 memcpy(&msg
, payload
->data
, sizeof(msg
));
2155 msg
.stream_id
= be64toh(msg
.stream_id
);
2157 stream
= stream_get_by_id(msg
.stream_id
);
2161 pthread_mutex_lock(&stream
->lock
);
2162 stream
->data_pending_check_done
= true;
2163 pthread_mutex_unlock(&stream
->lock
);
2165 DBG("Relay quiescent control pending flag set to %" PRIu64
, msg
.stream_id
);
2168 memset(&reply
, 0, sizeof(reply
));
2169 reply
.ret_code
= htobe32(LTTNG_OK
);
2170 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2171 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2172 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2184 * Initialize a data pending command. This means that a consumer is about
2185 * to ask for data pending for each stream it holds. Simply iterate over
2186 * all streams of a session and set the data_pending_check_done flag.
2188 * This command returns to the client a LTTNG_OK code.
2190 static int relay_begin_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2191 struct relay_connection
*conn
,
2192 const struct lttng_buffer_view
*payload
)
2196 struct lttng_ht_iter iter
;
2197 struct lttcomm_relayd_begin_data_pending msg
;
2198 struct lttcomm_relayd_generic_reply reply
;
2199 struct relay_stream
*stream
;
2204 DBG("Init streams for data pending");
2206 if (!conn
->session
|| !conn
->version_check_done
) {
2207 ERR("Trying to check for data before version check");
2209 goto end_no_session
;
2212 if (payload
->size
< sizeof(msg
)) {
2213 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2214 sizeof(msg
), payload
->size
);
2216 goto end_no_session
;
2218 memcpy(&msg
, payload
->data
, sizeof(msg
));
2219 msg
.session_id
= be64toh(msg
.session_id
);
2222 * Iterate over all streams to set the begin data pending flag.
2223 * For now, the streams are indexed by stream handle so we have
2224 * to iterate over all streams to find the one associated with
2225 * the right session_id.
2228 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2230 if (!stream_get(stream
)) {
2233 if (stream
->trace
->session
->id
== msg
.session_id
) {
2234 pthread_mutex_lock(&stream
->lock
);
2235 stream
->data_pending_check_done
= false;
2236 pthread_mutex_unlock(&stream
->lock
);
2237 DBG("Set begin data pending flag to stream %" PRIu64
,
2238 stream
->stream_handle
);
2244 memset(&reply
, 0, sizeof(reply
));
2245 /* All good, send back reply. */
2246 reply
.ret_code
= htobe32(LTTNG_OK
);
2248 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2249 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2250 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2262 * End data pending command. This will check, for a given session id, if
2263 * each stream associated with it has its data_pending_check_done flag
2264 * set. If not, this means that the client lost track of the stream but
2265 * the data is still being streamed on our side. In this case, we inform
2266 * the client that data is in flight.
2268 * Return to the client if there is data in flight or not with a ret_code.
2270 static int relay_end_data_pending(const struct lttcomm_relayd_hdr
*recv_hdr
,
2271 struct relay_connection
*conn
,
2272 const struct lttng_buffer_view
*payload
)
2276 struct lttng_ht_iter iter
;
2277 struct lttcomm_relayd_end_data_pending msg
;
2278 struct lttcomm_relayd_generic_reply reply
;
2279 struct relay_stream
*stream
;
2280 uint32_t is_data_inflight
= 0;
2282 DBG("End data pending command");
2284 if (!conn
->session
|| !conn
->version_check_done
) {
2285 ERR("Trying to check for data before version check");
2287 goto end_no_session
;
2290 if (payload
->size
< sizeof(msg
)) {
2291 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2292 sizeof(msg
), payload
->size
);
2294 goto end_no_session
;
2296 memcpy(&msg
, payload
->data
, sizeof(msg
));
2297 msg
.session_id
= be64toh(msg
.session_id
);
2300 * Iterate over all streams to see if the begin data pending
2304 cds_lfht_for_each_entry(relay_streams_ht
->ht
, &iter
.iter
, stream
,
2306 if (!stream_get(stream
)) {
2309 if (stream
->trace
->session
->id
!= msg
.session_id
) {
2313 pthread_mutex_lock(&stream
->lock
);
2314 if (!stream
->data_pending_check_done
) {
2315 uint64_t stream_seq
;
2317 if (session_streams_have_index(conn
->session
)) {
2319 * Ensure that both the index and stream data have been
2320 * flushed up to the requested point.
2322 stream_seq
= min(stream
->prev_data_seq
, stream
->prev_index_seq
);
2324 stream_seq
= stream
->prev_data_seq
;
2326 if (!stream
->closed
|| !(((int64_t) (stream_seq
- stream
->last_net_seq_num
)) >= 0)) {
2327 is_data_inflight
= 1;
2328 DBG("Data is still in flight for stream %" PRIu64
,
2329 stream
->stream_handle
);
2330 pthread_mutex_unlock(&stream
->lock
);
2335 pthread_mutex_unlock(&stream
->lock
);
2340 memset(&reply
, 0, sizeof(reply
));
2341 /* All good, send back reply. */
2342 reply
.ret_code
= htobe32(is_data_inflight
);
2344 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2345 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2346 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2358 * Receive an index for a specific stream.
2360 * Return 0 on success else a negative value.
2362 static int relay_recv_index(const struct lttcomm_relayd_hdr
*recv_hdr
,
2363 struct relay_connection
*conn
,
2364 const struct lttng_buffer_view
*payload
)
2368 struct relay_session
*session
= conn
->session
;
2369 struct lttcomm_relayd_index index_info
;
2370 struct lttcomm_relayd_generic_reply reply
;
2371 struct relay_stream
*stream
;
2376 DBG("Relay receiving index");
2378 if (!session
|| !conn
->version_check_done
) {
2379 ERR("Trying to close a stream before version check");
2381 goto end_no_session
;
2384 msg_len
= lttcomm_relayd_index_len(
2385 lttng_to_index_major(conn
->major
, conn
->minor
),
2386 lttng_to_index_minor(conn
->major
, conn
->minor
));
2387 if (payload
->size
< msg_len
) {
2388 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2389 msg_len
, payload
->size
);
2391 goto end_no_session
;
2393 memcpy(&index_info
, payload
->data
, msg_len
);
2394 index_info
.relay_stream_id
= be64toh(index_info
.relay_stream_id
);
2395 index_info
.net_seq_num
= be64toh(index_info
.net_seq_num
);
2396 index_info
.packet_size
= be64toh(index_info
.packet_size
);
2397 index_info
.content_size
= be64toh(index_info
.content_size
);
2398 index_info
.timestamp_begin
= be64toh(index_info
.timestamp_begin
);
2399 index_info
.timestamp_end
= be64toh(index_info
.timestamp_end
);
2400 index_info
.events_discarded
= be64toh(index_info
.events_discarded
);
2401 index_info
.stream_id
= be64toh(index_info
.stream_id
);
2403 if (conn
->minor
>= 8) {
2404 index_info
.stream_instance_id
=
2405 be64toh(index_info
.stream_instance_id
);
2406 index_info
.packet_seq_num
= be64toh(index_info
.packet_seq_num
);
2408 index_info
.stream_instance_id
= -1ULL;
2409 index_info
.packet_seq_num
= -1ULL;
2412 stream
= stream_get_by_id(index_info
.relay_stream_id
);
2414 ERR("stream_get_by_id not found");
2419 pthread_mutex_lock(&stream
->lock
);
2420 ret
= stream_add_index(stream
, &index_info
);
2421 pthread_mutex_unlock(&stream
->lock
);
2423 goto end_stream_put
;
2429 memset(&reply
, 0, sizeof(reply
));
2431 reply
.ret_code
= htobe32(LTTNG_ERR_UNK
);
2433 reply
.ret_code
= htobe32(LTTNG_OK
);
2435 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2436 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2437 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret
);
2446 * Receive the streams_sent message.
2448 * Return 0 on success else a negative value.
2450 static int relay_streams_sent(const struct lttcomm_relayd_hdr
*recv_hdr
,
2451 struct relay_connection
*conn
,
2452 const struct lttng_buffer_view
*payload
)
2456 struct lttcomm_relayd_generic_reply reply
;
2460 DBG("Relay receiving streams_sent");
2462 if (!conn
->session
|| !conn
->version_check_done
) {
2463 ERR("Trying to close a stream before version check");
2465 goto end_no_session
;
2469 * Publish every pending stream in the connection recv list which are
2470 * now ready to be used by the viewer.
2472 publish_connection_local_streams(conn
);
2474 memset(&reply
, 0, sizeof(reply
));
2475 reply
.ret_code
= htobe32(LTTNG_OK
);
2476 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
, sizeof(reply
), 0);
2477 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2478 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2490 static ssize_t
relay_unpack_rotate_streams_header(
2491 const struct lttng_buffer_view
*payload
,
2492 struct lttcomm_relayd_rotate_streams
*_rotate_streams
)
2494 struct lttcomm_relayd_rotate_streams rotate_streams
;
2496 * Set to the smallest version (packed) of `lttcomm_relayd_rotate_streams`.
2497 * This is the smallest version of this structure, but it can be larger;
2498 * this variable is updated once the proper size of the structure is known.
2500 * See comment at the declaration of this structure for more information.
2502 ssize_t header_len
= sizeof(struct lttcomm_relayd_rotate_streams_packed
);
2503 size_t expected_payload_size_no_padding
,
2504 expected_payload_size_3_bytes_padding
,
2505 expected_payload_size_7_bytes_padding
;
2507 if (payload
->size
< header_len
) {
2508 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2509 header_len
, payload
->size
);
2514 * Some versions incorrectly omitted the LTTNG_PACKED annotation on the
2515 * `new_chunk_id` optional field of struct lttcomm_relayd_rotate_streams.
2517 * We start by "unpacking" `stream_count` to figure out the padding length
2518 * emited by our peer.
2520 memcpy(&rotate_streams
.stream_count
, payload
->data
,
2521 sizeof(rotate_streams
.stream_count
));
2522 rotate_streams
= (typeof(rotate_streams
)) {
2523 .stream_count
= be32toh(rotate_streams
.stream_count
),
2527 * Payload size expected given the possible padding lengths in
2528 * `struct lttcomm_relayd_rotate_streams`.
2530 expected_payload_size_no_padding
= (rotate_streams
.stream_count
*
2531 sizeof(*rotate_streams
.rotation_positions
)) +
2532 sizeof(struct lttcomm_relayd_rotate_streams_packed
);
2533 expected_payload_size_3_bytes_padding
= (rotate_streams
.stream_count
*
2534 sizeof(*rotate_streams
.rotation_positions
)) +
2535 sizeof(struct lttcomm_relayd_rotate_streams_3_bytes_padding
);
2536 expected_payload_size_7_bytes_padding
= (rotate_streams
.stream_count
*
2537 sizeof(*rotate_streams
.rotation_positions
)) +
2538 sizeof(struct lttcomm_relayd_rotate_streams_7_bytes_padding
);
2540 if (payload
->size
== expected_payload_size_no_padding
) {
2541 struct lttcomm_relayd_rotate_streams_packed packed_rotate_streams
;
2544 * This handles cases where someone might build with
2545 * -fpack-struct or any other toolchain that wouldn't produce
2546 * padding to align `value`.
2548 DBG("Received `struct lttcomm_relayd_rotate_streams` with no padding");
2550 header_len
= sizeof(packed_rotate_streams
);
2551 memcpy(&packed_rotate_streams
, payload
->data
, header_len
);
2553 /* Unpack the packed structure to the natively-packed version. */
2554 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2555 .stream_count
= be32toh(packed_rotate_streams
.stream_count
),
2556 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2557 .is_set
= !!packed_rotate_streams
.new_chunk_id
.is_set
,
2558 .value
= be64toh(packed_rotate_streams
.new_chunk_id
.value
),
2561 } else if (payload
->size
== expected_payload_size_3_bytes_padding
) {
2562 struct lttcomm_relayd_rotate_streams_3_bytes_padding padded_rotate_streams
;
2564 DBG("Received `struct lttcomm_relayd_rotate_streams` with 3 bytes of padding (4-byte aligned peer)");
2566 header_len
= sizeof(padded_rotate_streams
);
2567 memcpy(&padded_rotate_streams
, payload
->data
, header_len
);
2569 /* Unpack the 3-byte padded structure to the natively-packed version. */
2570 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2571 .stream_count
= be32toh(padded_rotate_streams
.stream_count
),
2572 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2573 .is_set
= !!padded_rotate_streams
.new_chunk_id
.is_set
,
2574 .value
= be64toh(padded_rotate_streams
.new_chunk_id
.value
),
2577 } else if (payload
->size
== expected_payload_size_7_bytes_padding
) {
2578 struct lttcomm_relayd_rotate_streams_7_bytes_padding padded_rotate_streams
;
2580 DBG("Received `struct lttcomm_relayd_rotate_streams` with 7 bytes of padding (8-byte aligned peer)");
2582 header_len
= sizeof(padded_rotate_streams
);
2583 memcpy(&padded_rotate_streams
, payload
->data
, header_len
);
2585 /* Unpack the 7-byte padded structure to the natively-packed version. */
2586 *_rotate_streams
= (typeof(*_rotate_streams
)) {
2587 .stream_count
= be32toh(padded_rotate_streams
.stream_count
),
2588 .new_chunk_id
= (typeof(_rotate_streams
->new_chunk_id
)) {
2589 .is_set
= !!padded_rotate_streams
.new_chunk_id
.is_set
,
2590 .value
= be64toh(padded_rotate_streams
.new_chunk_id
.value
),
2594 header_len
= sizeof(padded_rotate_streams
);
2596 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected %zu, %zu or %zu bytes, got %zu bytes",
2597 expected_payload_size_no_padding
,
2598 expected_payload_size_3_bytes_padding
,
2599 expected_payload_size_7_bytes_padding
,
2610 * relay_rotate_session_stream: rotate a stream to a new tracefile for the
2611 * session rotation feature (not the tracefile rotation feature).
2613 static int relay_rotate_session_streams(
2614 const struct lttcomm_relayd_hdr
*recv_hdr
,
2615 struct relay_connection
*conn
,
2616 const struct lttng_buffer_view
*payload
)
2621 enum lttng_error_code reply_code
= LTTNG_ERR_UNK
;
2622 struct relay_session
*session
= conn
->session
;
2623 struct lttcomm_relayd_rotate_streams rotate_streams
;
2624 struct lttcomm_relayd_generic_reply reply
= {};
2625 struct relay_stream
*stream
= NULL
;
2626 struct lttng_trace_chunk
*next_trace_chunk
= NULL
;
2627 struct lttng_buffer_view stream_positions
;
2628 char chunk_id_buf
[MAX_INT_DEC_LEN(uint64_t)];
2629 const char *chunk_id_str
= "none";
2632 if (!session
|| !conn
->version_check_done
) {
2633 ERR("Trying to rotate a stream before version check");
2638 if (session
->major
== 2 && session
->minor
< 11) {
2639 ERR("Unsupported feature before 2.11");
2644 header_len
= relay_unpack_rotate_streams_header(payload
, &rotate_streams
);
2645 if (header_len
< 0) {
2650 if (rotate_streams
.new_chunk_id
.is_set
) {
2652 * Retrieve the trace chunk the stream must transition to. As
2653 * per the protocol, this chunk should have been created
2654 * before this command is received.
2656 next_trace_chunk
= sessiond_trace_chunk_registry_get_chunk(
2657 sessiond_trace_chunk_registry
,
2658 session
->sessiond_uuid
,
2659 conn
->session
->id_sessiond
.is_set
?
2660 conn
->session
->id_sessiond
.value
:
2662 rotate_streams
.new_chunk_id
.value
);
2663 if (!next_trace_chunk
) {
2664 char uuid_str
[LTTNG_UUID_STR_LEN
];
2666 lttng_uuid_to_str(session
->sessiond_uuid
, uuid_str
);
2667 ERR("Unknown next trace chunk in ROTATE_STREAMS command: sessiond_uuid = {%s}, session_id = %" PRIu64
2668 ", trace_chunk_id = %" PRIu64
,
2669 uuid_str
, session
->id
,
2670 rotate_streams
.new_chunk_id
.value
);
2671 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2676 ret
= snprintf(chunk_id_buf
, sizeof(chunk_id_buf
), "%" PRIu64
,
2677 rotate_streams
.new_chunk_id
.value
);
2678 if (ret
< 0 || ret
>= sizeof(chunk_id_buf
)) {
2679 chunk_id_str
= "formatting error";
2681 chunk_id_str
= chunk_id_buf
;
2685 DBG("Rotate %" PRIu32
" streams of session \"%s\" to chunk \"%s\"",
2686 rotate_streams
.stream_count
, session
->session_name
,
2689 stream_positions
= lttng_buffer_view_from_view(payload
,
2691 if (!stream_positions
.data
||
2692 stream_positions
.size
<
2693 (rotate_streams
.stream_count
*
2694 sizeof(struct lttcomm_relayd_stream_rotation_position
))) {
2695 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2700 for (i
= 0; i
< rotate_streams
.stream_count
; i
++) {
2701 struct lttcomm_relayd_stream_rotation_position
*position_comm
=
2702 &((typeof(position_comm
)) stream_positions
.data
)[i
];
2703 const struct lttcomm_relayd_stream_rotation_position pos
= {
2704 .stream_id
= be64toh(position_comm
->stream_id
),
2705 .rotate_at_seq_num
= be64toh(
2706 position_comm
->rotate_at_seq_num
),
2709 stream
= stream_get_by_id(pos
.stream_id
);
2711 reply_code
= LTTNG_ERR_INVALID
;
2716 pthread_mutex_lock(&stream
->lock
);
2717 ret
= stream_set_pending_rotation(stream
, next_trace_chunk
,
2718 pos
.rotate_at_seq_num
);
2719 pthread_mutex_unlock(&stream
->lock
);
2721 reply_code
= LTTNG_ERR_FILE_CREATION_ERROR
;
2729 reply_code
= LTTNG_OK
;
2736 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2737 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
, &reply
,
2738 sizeof(struct lttcomm_relayd_generic_reply
), 0);
2739 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2740 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2745 lttng_trace_chunk_put(next_trace_chunk
);
2750 * relay_create_trace_chunk: create a new trace chunk
2752 static int relay_create_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2753 struct relay_connection
*conn
,
2754 const struct lttng_buffer_view
*payload
)
2758 struct relay_session
*session
= conn
->session
;
2759 struct lttcomm_relayd_create_trace_chunk
*msg
;
2760 struct lttcomm_relayd_generic_reply reply
= {};
2761 struct lttng_buffer_view header_view
;
2762 struct lttng_buffer_view chunk_name_view
;
2763 struct lttng_trace_chunk
*chunk
= NULL
, *published_chunk
= NULL
;
2764 enum lttng_error_code reply_code
= LTTNG_OK
;
2765 enum lttng_trace_chunk_status chunk_status
;
2766 const char *new_path
;
2768 if (!session
|| !conn
->version_check_done
) {
2769 ERR("Trying to create a trace chunk before version check");
2774 if (session
->major
== 2 && session
->minor
< 11) {
2775 ERR("Chunk creation command is unsupported before 2.11");
2780 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2781 if (!header_view
.data
) {
2782 ERR("Failed to receive payload of chunk creation command");
2787 /* Convert to host endianness. */
2788 msg
= (typeof(msg
)) header_view
.data
;
2789 msg
->chunk_id
= be64toh(msg
->chunk_id
);
2790 msg
->creation_timestamp
= be64toh(msg
->creation_timestamp
);
2791 msg
->override_name_length
= be32toh(msg
->override_name_length
);
2793 pthread_mutex_lock(&conn
->session
->lock
);
2794 session
->ongoing_rotation
= true;
2795 if (session
->current_trace_chunk
&&
2796 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
2797 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
2798 DEFAULT_CHUNK_TMP_OLD_DIRECTORY
);
2799 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2800 ERR("Failed to rename old chunk");
2802 reply_code
= LTTNG_ERR_UNK
;
2806 if (!session
->current_trace_chunk
) {
2807 if (!session
->has_rotated
) {
2813 new_path
= DEFAULT_CHUNK_TMP_NEW_DIRECTORY
;
2815 chunk
= lttng_trace_chunk_create(
2816 msg
->chunk_id
, msg
->creation_timestamp
, new_path
);
2818 ERR("Failed to create trace chunk in trace chunk creation command");
2820 reply_code
= LTTNG_ERR_NOMEM
;
2823 lttng_trace_chunk_set_fd_tracker(chunk
, the_fd_tracker
);
2825 if (msg
->override_name_length
) {
2828 chunk_name_view
= lttng_buffer_view_from_view(payload
,
2830 msg
->override_name_length
);
2831 name
= chunk_name_view
.data
;
2832 if (!name
|| name
[msg
->override_name_length
- 1]) {
2833 ERR("Failed to receive payload of chunk creation command");
2835 reply_code
= LTTNG_ERR_INVALID
;
2839 chunk_status
= lttng_trace_chunk_override_name(
2840 chunk
, chunk_name_view
.data
);
2841 switch (chunk_status
) {
2842 case LTTNG_TRACE_CHUNK_STATUS_OK
:
2844 case LTTNG_TRACE_CHUNK_STATUS_INVALID_ARGUMENT
:
2845 ERR("Failed to set the name of new trace chunk in trace chunk creation command (invalid name)");
2846 reply_code
= LTTNG_ERR_INVALID
;
2850 ERR("Failed to set the name of new trace chunk in trace chunk creation command (unknown error)");
2851 reply_code
= LTTNG_ERR_UNK
;
2857 chunk_status
= lttng_trace_chunk_set_credentials_current_user(chunk
);
2858 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2859 reply_code
= LTTNG_ERR_UNK
;
2864 assert(conn
->session
->output_directory
);
2865 chunk_status
= lttng_trace_chunk_set_as_owner(chunk
,
2866 conn
->session
->output_directory
);
2867 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
2868 reply_code
= LTTNG_ERR_UNK
;
2873 published_chunk
= sessiond_trace_chunk_registry_publish_chunk(
2874 sessiond_trace_chunk_registry
,
2875 conn
->session
->sessiond_uuid
,
2876 conn
->session
->id_sessiond
.is_set
?
2877 conn
->session
->id_sessiond
.value
:
2880 if (!published_chunk
) {
2881 char uuid_str
[LTTNG_UUID_STR_LEN
];
2883 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2884 ERR("Failed to publish chunk: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2889 reply_code
= LTTNG_ERR_NOMEM
;
2893 if (conn
->session
->pending_closure_trace_chunk
) {
2895 * Invalid; this means a second create_trace_chunk command was
2896 * received before a close_trace_chunk.
2898 ERR("Invalid trace chunk close command received; a trace chunk is already waiting for a trace chunk close command");
2899 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
2903 conn
->session
->pending_closure_trace_chunk
=
2904 conn
->session
->current_trace_chunk
;
2905 conn
->session
->current_trace_chunk
= published_chunk
;
2906 published_chunk
= NULL
;
2907 if (!conn
->session
->pending_closure_trace_chunk
) {
2908 session
->ongoing_rotation
= false;
2911 pthread_mutex_unlock(&conn
->session
->lock
);
2912 reply
.ret_code
= htobe32((uint32_t) reply_code
);
2913 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
2915 sizeof(struct lttcomm_relayd_generic_reply
),
2917 if (send_ret
< (ssize_t
) sizeof(reply
)) {
2918 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
2923 lttng_trace_chunk_put(chunk
);
2924 lttng_trace_chunk_put(published_chunk
);
2929 * relay_close_trace_chunk: close a trace chunk
2931 static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr
*recv_hdr
,
2932 struct relay_connection
*conn
,
2933 const struct lttng_buffer_view
*payload
)
2935 int ret
= 0, buf_ret
;
2937 struct relay_session
*session
= conn
->session
;
2938 struct lttcomm_relayd_close_trace_chunk
*msg
;
2939 struct lttcomm_relayd_close_trace_chunk_reply reply
= {};
2940 struct lttng_buffer_view header_view
;
2941 struct lttng_trace_chunk
*chunk
= NULL
;
2942 enum lttng_error_code reply_code
= LTTNG_OK
;
2943 enum lttng_trace_chunk_status chunk_status
;
2945 LTTNG_OPTIONAL(enum lttng_trace_chunk_command_type
) close_command
= {};
2946 time_t close_timestamp
;
2947 char closed_trace_chunk_path
[LTTNG_PATH_MAX
];
2948 size_t path_length
= 0;
2949 const char *chunk_name
= NULL
;
2950 struct lttng_dynamic_buffer reply_payload
;
2951 const char *new_path
;
2953 lttng_dynamic_buffer_init(&reply_payload
);
2955 if (!session
|| !conn
->version_check_done
) {
2956 ERR("Trying to close a trace chunk before version check");
2961 if (session
->major
== 2 && session
->minor
< 11) {
2962 ERR("Chunk close command is unsupported before 2.11");
2967 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
2968 if (!header_view
.data
) {
2969 ERR("Failed to receive payload of chunk close command");
2974 /* Convert to host endianness. */
2975 msg
= (typeof(msg
)) header_view
.data
;
2976 chunk_id
= be64toh(msg
->chunk_id
);
2977 close_timestamp
= (time_t) be64toh(msg
->close_timestamp
);
2978 close_command
= (typeof(close_command
)){
2979 .value
= be32toh(msg
->close_command
.value
),
2980 .is_set
= msg
->close_command
.is_set
,
2983 chunk
= sessiond_trace_chunk_registry_get_chunk(
2984 sessiond_trace_chunk_registry
,
2985 conn
->session
->sessiond_uuid
,
2986 conn
->session
->id_sessiond
.is_set
?
2987 conn
->session
->id_sessiond
.value
:
2991 char uuid_str
[LTTNG_UUID_STR_LEN
];
2993 lttng_uuid_to_str(conn
->session
->sessiond_uuid
, uuid_str
);
2994 ERR("Failed to find chunk to close: sessiond_uuid = %s, session_id = %" PRIu64
", chunk_id = %" PRIu64
,
2999 reply_code
= LTTNG_ERR_NOMEM
;
3003 pthread_mutex_lock(&session
->lock
);
3004 if (close_command
.is_set
&&
3005 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
) {
3007 * Clear command. It is a protocol error to ask for a
3008 * clear on a relay which does not allow it. Querying
3009 * the configuration allows figuring out whether
3010 * clearing is allowed before doing the clear.
3012 if (!opt_allow_clear
) {
3014 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3015 goto end_unlock_session
;
3018 if (session
->pending_closure_trace_chunk
&&
3019 session
->pending_closure_trace_chunk
!= chunk
) {
3020 ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
3021 session
->session_name
);
3022 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3024 goto end_unlock_session
;
3027 if (session
->current_trace_chunk
&& session
->current_trace_chunk
!= chunk
&&
3028 !lttng_trace_chunk_get_name_overridden(session
->current_trace_chunk
)) {
3029 if (close_command
.is_set
&&
3030 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE
&&
3031 !session
->has_rotated
) {
3032 /* New chunk stays in session output directory. */
3035 /* Use chunk name for new chunk. */
3038 /* Rename new chunk path. */
3039 chunk_status
= lttng_trace_chunk_rename_path(session
->current_trace_chunk
,
3041 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3043 goto end_unlock_session
;
3045 session
->ongoing_rotation
= false;
3047 if ((!close_command
.is_set
||
3048 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_NO_OPERATION
) &&
3049 !lttng_trace_chunk_get_name_overridden(chunk
)) {
3050 const char *old_path
;
3052 if (!session
->has_rotated
) {
3057 /* We need to move back the .tmp_old_chunk to its rightful place. */
3058 chunk_status
= lttng_trace_chunk_rename_path(chunk
, old_path
);
3059 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3061 goto end_unlock_session
;
3064 chunk_status
= lttng_trace_chunk_set_close_timestamp(
3065 chunk
, close_timestamp
);
3066 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3067 ERR("Failed to set trace chunk close timestamp");
3069 reply_code
= LTTNG_ERR_UNK
;
3070 goto end_unlock_session
;
3073 if (close_command
.is_set
) {
3074 chunk_status
= lttng_trace_chunk_set_close_command(
3075 chunk
, close_command
.value
);
3076 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3078 reply_code
= LTTNG_ERR_INVALID
;
3079 goto end_unlock_session
;
3082 chunk_status
= lttng_trace_chunk_get_name(chunk
, &chunk_name
, NULL
);
3083 if (chunk_status
!= LTTNG_TRACE_CHUNK_STATUS_OK
) {
3084 ERR("Failed to get chunk name");
3086 reply_code
= LTTNG_ERR_UNK
;
3087 goto end_unlock_session
;
3089 if (!session
->has_rotated
&& !session
->snapshot
) {
3090 ret
= lttng_strncpy(closed_trace_chunk_path
,
3091 session
->output_path
,
3092 sizeof(closed_trace_chunk_path
));
3094 ERR("Failed to send trace chunk path: path length of %zu bytes exceeds the maximal allowed length of %zu bytes",
3095 strlen(session
->output_path
),
3096 sizeof(closed_trace_chunk_path
));
3097 reply_code
= LTTNG_ERR_NOMEM
;
3099 goto end_unlock_session
;
3102 if (session
->snapshot
) {
3103 ret
= snprintf(closed_trace_chunk_path
,
3104 sizeof(closed_trace_chunk_path
),
3105 "%s/%s", session
->output_path
,
3108 ret
= snprintf(closed_trace_chunk_path
,
3109 sizeof(closed_trace_chunk_path
),
3110 "%s/" DEFAULT_ARCHIVED_TRACE_CHUNKS_DIRECTORY
3112 session
->output_path
, chunk_name
);
3114 if (ret
< 0 || ret
== sizeof(closed_trace_chunk_path
)) {
3115 ERR("Failed to format closed trace chunk resulting path");
3116 reply_code
= ret
< 0 ? LTTNG_ERR_UNK
: LTTNG_ERR_NOMEM
;
3118 goto end_unlock_session
;
3121 if (close_command
.is_set
&&
3122 close_command
.value
== LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED
) {
3123 session
->has_rotated
= true;
3125 DBG("Reply chunk path on close: %s", closed_trace_chunk_path
);
3126 path_length
= strlen(closed_trace_chunk_path
) + 1;
3127 if (path_length
> UINT32_MAX
) {
3128 ERR("Closed trace chunk path exceeds the maximal length allowed by the protocol");
3130 reply_code
= LTTNG_ERR_INVALID_PROTOCOL
;
3131 goto end_unlock_session
;
3134 if (session
->current_trace_chunk
== chunk
) {
3136 * After a trace chunk close command, no new streams
3137 * referencing the chunk may be created. Hence, on the
3138 * event that no new trace chunk have been created for
3139 * the session, the reference to the current trace chunk
3140 * is released in order to allow it to be reclaimed when
3141 * the last stream releases its reference to it.
3143 lttng_trace_chunk_put(session
->current_trace_chunk
);
3144 session
->current_trace_chunk
= NULL
;
3146 lttng_trace_chunk_put(session
->pending_closure_trace_chunk
);
3147 session
->pending_closure_trace_chunk
= NULL
;
3149 pthread_mutex_unlock(&session
->lock
);
3152 reply
.generic
.ret_code
= htobe32((uint32_t) reply_code
);
3153 reply
.path_length
= htobe32((uint32_t) path_length
);
3154 buf_ret
= lttng_dynamic_buffer_append(
3155 &reply_payload
, &reply
, sizeof(reply
));
3157 ERR("Failed to append \"close trace chunk\" command reply header to payload buffer");
3161 if (reply_code
== LTTNG_OK
) {
3162 buf_ret
= lttng_dynamic_buffer_append(&reply_payload
,
3163 closed_trace_chunk_path
, path_length
);
3165 ERR("Failed to append \"close trace chunk\" command reply path to payload buffer");
3170 send_ret
= conn
->sock
->ops
->sendmsg(conn
->sock
,
3174 if (send_ret
< reply_payload
.size
) {
3175 ERR("Failed to send \"close trace chunk\" command reply of %zu bytes (ret = %zd)",
3176 reply_payload
.size
, send_ret
);
3181 lttng_trace_chunk_put(chunk
);
3182 lttng_dynamic_buffer_reset(&reply_payload
);
3187 * relay_trace_chunk_exists: check if a trace chunk exists
3189 static int relay_trace_chunk_exists(const struct lttcomm_relayd_hdr
*recv_hdr
,
3190 struct relay_connection
*conn
,
3191 const struct lttng_buffer_view
*payload
)
3195 struct relay_session
*session
= conn
->session
;
3196 struct lttcomm_relayd_trace_chunk_exists
*msg
;
3197 struct lttcomm_relayd_trace_chunk_exists_reply reply
= {};
3198 struct lttng_buffer_view header_view
;
3202 if (!session
|| !conn
->version_check_done
) {
3203 ERR("Trying to close a trace chunk before version check");
3208 if (session
->major
== 2 && session
->minor
< 11) {
3209 ERR("Chunk close command is unsupported before 2.11");
3214 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3215 if (!header_view
.data
) {
3216 ERR("Failed to receive payload of chunk close command");
3221 /* Convert to host endianness. */
3222 msg
= (typeof(msg
)) header_view
.data
;
3223 chunk_id
= be64toh(msg
->chunk_id
);
3225 ret
= sessiond_trace_chunk_registry_chunk_exists(
3226 sessiond_trace_chunk_registry
,
3227 conn
->session
->sessiond_uuid
,
3229 chunk_id
, &chunk_exists
);
3231 * If ret is not 0, send the reply and report the error to the caller.
3232 * It is a protocol (or internal) error and the session/connection
3233 * should be torn down.
3235 reply
= (typeof(reply
)){
3236 .generic
.ret_code
= htobe32((uint32_t)
3237 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3238 .trace_chunk_exists
= ret
== 0 ? chunk_exists
: 0,
3240 send_ret
= conn
->sock
->ops
->sendmsg(
3241 conn
->sock
, &reply
, sizeof(reply
), 0);
3242 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3243 ERR("Failed to send \"create trace chunk\" command reply (ret = %zd)",
3252 * relay_get_configuration: query whether feature is available
3254 static int relay_get_configuration(const struct lttcomm_relayd_hdr
*recv_hdr
,
3255 struct relay_connection
*conn
,
3256 const struct lttng_buffer_view
*payload
)
3260 struct lttcomm_relayd_get_configuration
*msg
;
3261 struct lttcomm_relayd_get_configuration_reply reply
= {};
3262 struct lttng_buffer_view header_view
;
3263 uint64_t query_flags
= 0;
3264 uint64_t result_flags
= 0;
3266 header_view
= lttng_buffer_view_from_view(payload
, 0, sizeof(*msg
));
3267 if (!header_view
.data
) {
3268 ERR("Failed to receive payload of chunk close command");
3273 /* Convert to host endianness. */
3274 msg
= (typeof(msg
)) header_view
.data
;
3275 query_flags
= be64toh(msg
->query_flags
);
3278 ret
= LTTNG_ERR_INVALID_PROTOCOL
;
3281 if (opt_allow_clear
) {
3282 result_flags
|= LTTCOMM_RELAYD_CONFIGURATION_FLAG_CLEAR_ALLOWED
;
3286 reply
= (typeof(reply
)){
3287 .generic
.ret_code
= htobe32((uint32_t)
3288 (ret
== 0 ? LTTNG_OK
: LTTNG_ERR_INVALID_PROTOCOL
)),
3289 .relayd_configuration_flags
= htobe64(result_flags
),
3291 send_ret
= conn
->sock
->ops
->sendmsg(
3292 conn
->sock
, &reply
, sizeof(reply
), 0);
3293 if (send_ret
< (ssize_t
) sizeof(reply
)) {
3294 ERR("Failed to send \"get configuration\" command reply (ret = %zd)",
3302 #define DBG_CMD(cmd_name, conn) \
3303 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
3305 static int relay_process_control_command(struct relay_connection
*conn
,
3306 const struct lttcomm_relayd_hdr
*header
,
3307 const struct lttng_buffer_view
*payload
)
3311 switch (header
->cmd
) {
3312 case RELAYD_CREATE_SESSION
:
3313 DBG_CMD("RELAYD_CREATE_SESSION", conn
);
3314 ret
= relay_create_session(header
, conn
, payload
);
3316 case RELAYD_ADD_STREAM
:
3317 DBG_CMD("RELAYD_ADD_STREAM", conn
);
3318 ret
= relay_add_stream(header
, conn
, payload
);
3320 case RELAYD_START_DATA
:
3321 DBG_CMD("RELAYD_START_DATA", conn
);
3322 ret
= relay_start(header
, conn
, payload
);
3324 case RELAYD_SEND_METADATA
:
3325 DBG_CMD("RELAYD_SEND_METADATA", conn
);
3326 ret
= relay_recv_metadata(header
, conn
, payload
);
3328 case RELAYD_VERSION
:
3329 DBG_CMD("RELAYD_VERSION", conn
);
3330 ret
= relay_send_version(header
, conn
, payload
);
3332 case RELAYD_CLOSE_STREAM
:
3333 DBG_CMD("RELAYD_CLOSE_STREAM", conn
);
3334 ret
= relay_close_stream(header
, conn
, payload
);
3336 case RELAYD_DATA_PENDING
:
3337 DBG_CMD("RELAYD_DATA_PENDING", conn
);
3338 ret
= relay_data_pending(header
, conn
, payload
);
3340 case RELAYD_QUIESCENT_CONTROL
:
3341 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn
);
3342 ret
= relay_quiescent_control(header
, conn
, payload
);
3344 case RELAYD_BEGIN_DATA_PENDING
:
3345 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn
);
3346 ret
= relay_begin_data_pending(header
, conn
, payload
);
3348 case RELAYD_END_DATA_PENDING
:
3349 DBG_CMD("RELAYD_END_DATA_PENDING", conn
);
3350 ret
= relay_end_data_pending(header
, conn
, payload
);
3352 case RELAYD_SEND_INDEX
:
3353 DBG_CMD("RELAYD_SEND_INDEX", conn
);
3354 ret
= relay_recv_index(header
, conn
, payload
);
3356 case RELAYD_STREAMS_SENT
:
3357 DBG_CMD("RELAYD_STREAMS_SENT", conn
);
3358 ret
= relay_streams_sent(header
, conn
, payload
);
3360 case RELAYD_RESET_METADATA
:
3361 DBG_CMD("RELAYD_RESET_METADATA", conn
);
3362 ret
= relay_reset_metadata(header
, conn
, payload
);
3364 case RELAYD_ROTATE_STREAMS
:
3365 DBG_CMD("RELAYD_ROTATE_STREAMS", conn
);
3366 ret
= relay_rotate_session_streams(header
, conn
, payload
);
3368 case RELAYD_CREATE_TRACE_CHUNK
:
3369 DBG_CMD("RELAYD_CREATE_TRACE_CHUNK", conn
);
3370 ret
= relay_create_trace_chunk(header
, conn
, payload
);
3372 case RELAYD_CLOSE_TRACE_CHUNK
:
3373 DBG_CMD("RELAYD_CLOSE_TRACE_CHUNK", conn
);
3374 ret
= relay_close_trace_chunk(header
, conn
, payload
);
3376 case RELAYD_TRACE_CHUNK_EXISTS
:
3377 DBG_CMD("RELAYD_TRACE_CHUNK_EXISTS", conn
);
3378 ret
= relay_trace_chunk_exists(header
, conn
, payload
);
3380 case RELAYD_GET_CONFIGURATION
:
3381 DBG_CMD("RELAYD_GET_CONFIGURATION", conn
);
3382 ret
= relay_get_configuration(header
, conn
, payload
);
3384 case RELAYD_UPDATE_SYNC_INFO
:
3386 ERR("Received unknown command (%u)", header
->cmd
);
3387 relay_unknown_command(conn
);
3396 static enum relay_connection_status
relay_process_control_receive_payload(
3397 struct relay_connection
*conn
)
3400 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3401 struct lttng_dynamic_buffer
*reception_buffer
=
3402 &conn
->protocol
.ctrl
.reception_buffer
;
3403 struct ctrl_connection_state_receive_payload
*state
=
3404 &conn
->protocol
.ctrl
.state
.receive_payload
;
3405 struct lttng_buffer_view payload_view
;
3407 if (state
->left_to_receive
== 0) {
3408 /* Short-circuit for payload-less commands. */
3409 goto reception_complete
;
3412 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3413 reception_buffer
->data
+ state
->received
,
3414 state
->left_to_receive
, MSG_DONTWAIT
);
3416 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3417 PERROR("Unable to receive command payload on sock %d",
3419 status
= RELAY_CONNECTION_STATUS_ERROR
;
3422 } else if (ret
== 0) {
3423 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3424 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3429 assert(ret
<= state
->left_to_receive
);
3431 state
->left_to_receive
-= ret
;
3432 state
->received
+= ret
;
3434 if (state
->left_to_receive
> 0) {
3436 * Can't transition to the protocol's next state, wait to
3437 * receive the rest of the header.
3439 DBG3("Partial reception of control connection protocol payload (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3440 state
->received
, state
->left_to_receive
,
3446 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64
" bytes",
3447 conn
->sock
->fd
, state
->received
);
3449 * The payload required to process the command has been received.
3450 * A view to the reception buffer is forwarded to the various
3451 * commands and the state of the control is reset on success.
3453 * Commands are responsible for sending their reply to the peer.
3455 payload_view
= lttng_buffer_view_from_dynamic_buffer(reception_buffer
,
3457 ret
= relay_process_control_command(conn
,
3458 &state
->header
, &payload_view
);
3460 status
= RELAY_CONNECTION_STATUS_ERROR
;
3464 ret
= connection_reset_protocol_state(conn
);
3466 status
= RELAY_CONNECTION_STATUS_ERROR
;
3472 static enum relay_connection_status
relay_process_control_receive_header(
3473 struct relay_connection
*conn
)
3476 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3477 struct lttcomm_relayd_hdr header
;
3478 struct lttng_dynamic_buffer
*reception_buffer
=
3479 &conn
->protocol
.ctrl
.reception_buffer
;
3480 struct ctrl_connection_state_receive_header
*state
=
3481 &conn
->protocol
.ctrl
.state
.receive_header
;
3483 assert(state
->left_to_receive
!= 0);
3485 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3486 reception_buffer
->data
+ state
->received
,
3487 state
->left_to_receive
, MSG_DONTWAIT
);
3489 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3490 PERROR("Unable to receive control command header on sock %d",
3492 status
= RELAY_CONNECTION_STATUS_ERROR
;
3495 } else if (ret
== 0) {
3496 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3497 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3502 assert(ret
<= state
->left_to_receive
);
3504 state
->left_to_receive
-= ret
;
3505 state
->received
+= ret
;
3507 if (state
->left_to_receive
> 0) {
3509 * Can't transition to the protocol's next state, wait to
3510 * receive the rest of the header.
3512 DBG3("Partial reception of control connection protocol header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3513 state
->received
, state
->left_to_receive
,
3518 /* Transition to next state: receiving the command's payload. */
3519 conn
->protocol
.ctrl
.state_id
=
3520 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3521 memcpy(&header
, reception_buffer
->data
, sizeof(header
));
3522 header
.circuit_id
= be64toh(header
.circuit_id
);
3523 header
.data_size
= be64toh(header
.data_size
);
3524 header
.cmd
= be32toh(header
.cmd
);
3525 header
.cmd_version
= be32toh(header
.cmd_version
);
3526 memcpy(&conn
->protocol
.ctrl
.state
.receive_payload
.header
,
3527 &header
, sizeof(header
));
3529 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32
", cmd_version = %" PRIu32
", payload size = %" PRIu64
" bytes",
3530 conn
->sock
->fd
, header
.cmd
, header
.cmd_version
,
3533 if (header
.data_size
> DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE
) {
3534 ERR("Command header indicates a payload (%" PRIu64
" bytes) that exceeds the maximal payload size allowed on a control connection.",
3536 status
= RELAY_CONNECTION_STATUS_ERROR
;
3540 conn
->protocol
.ctrl
.state
.receive_payload
.left_to_receive
=
3542 conn
->protocol
.ctrl
.state
.receive_payload
.received
= 0;
3543 ret
= lttng_dynamic_buffer_set_size(reception_buffer
,
3546 status
= RELAY_CONNECTION_STATUS_ERROR
;
3550 if (header
.data_size
== 0) {
3552 * Manually invoke the next state as the poll loop
3553 * will not wake-up to allow us to proceed further.
3555 status
= relay_process_control_receive_payload(conn
);
3562 * Process the commands received on the control socket
3564 static enum relay_connection_status
relay_process_control(
3565 struct relay_connection
*conn
)
3567 enum relay_connection_status status
;
3569 switch (conn
->protocol
.ctrl
.state_id
) {
3570 case CTRL_CONNECTION_STATE_RECEIVE_HEADER
:
3571 status
= relay_process_control_receive_header(conn
);
3573 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3574 status
= relay_process_control_receive_payload(conn
);
3577 ERR("Unknown control connection protocol state encountered.");
3584 static enum relay_connection_status
relay_process_data_receive_header(
3585 struct relay_connection
*conn
)
3588 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3589 struct data_connection_state_receive_header
*state
=
3590 &conn
->protocol
.data
.state
.receive_header
;
3591 struct lttcomm_relayd_data_hdr header
;
3592 struct relay_stream
*stream
;
3594 assert(state
->left_to_receive
!= 0);
3596 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
,
3597 state
->header_reception_buffer
+ state
->received
,
3598 state
->left_to_receive
, MSG_DONTWAIT
);
3600 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3601 PERROR("Unable to receive data header on sock %d", conn
->sock
->fd
);
3602 status
= RELAY_CONNECTION_STATUS_ERROR
;
3605 } else if (ret
== 0) {
3606 /* Orderly shutdown. Not necessary to print an error. */
3607 DBG("Socket %d performed an orderly shutdown (received EOF)", conn
->sock
->fd
);
3608 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3613 assert(ret
<= state
->left_to_receive
);
3615 state
->left_to_receive
-= ret
;
3616 state
->received
+= ret
;
3618 if (state
->left_to_receive
> 0) {
3620 * Can't transition to the protocol's next state, wait to
3621 * receive the rest of the header.
3623 DBG3("Partial reception of data connection header (received %" PRIu64
" bytes, %" PRIu64
" bytes left to receive, fd = %i)",
3624 state
->received
, state
->left_to_receive
,
3629 /* Transition to next state: receiving the payload. */
3630 conn
->protocol
.data
.state_id
= DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
;
3632 memcpy(&header
, state
->header_reception_buffer
, sizeof(header
));
3633 header
.circuit_id
= be64toh(header
.circuit_id
);
3634 header
.stream_id
= be64toh(header
.stream_id
);
3635 header
.data_size
= be32toh(header
.data_size
);
3636 header
.net_seq_num
= be64toh(header
.net_seq_num
);
3637 header
.padding_size
= be32toh(header
.padding_size
);
3638 memcpy(&conn
->protocol
.data
.state
.receive_payload
.header
, &header
, sizeof(header
));
3640 conn
->protocol
.data
.state
.receive_payload
.left_to_receive
=
3642 conn
->protocol
.data
.state
.receive_payload
.received
= 0;
3643 conn
->protocol
.data
.state
.receive_payload
.rotate_index
= false;
3645 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64
", stream_id = %" PRIu64
", data_size = %" PRIu32
", net_seq_num = %" PRIu64
", padding_size = %" PRIu32
,
3646 conn
->sock
->fd
, header
.circuit_id
,
3647 header
.stream_id
, header
.data_size
,
3648 header
.net_seq_num
, header
.padding_size
);
3650 stream
= stream_get_by_id(header
.stream_id
);
3652 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64
,
3654 /* Protocol error. */
3655 status
= RELAY_CONNECTION_STATUS_ERROR
;
3659 pthread_mutex_lock(&stream
->lock
);
3660 /* Prepare stream for the reception of a new packet. */
3661 ret
= stream_init_packet(stream
, header
.data_size
,
3662 &conn
->protocol
.data
.state
.receive_payload
.rotate_index
);
3663 pthread_mutex_unlock(&stream
->lock
);
3665 ERR("Failed to rotate stream output file");
3666 status
= RELAY_CONNECTION_STATUS_ERROR
;
3667 goto end_stream_unlock
;
3676 static enum relay_connection_status
relay_process_data_receive_payload(
3677 struct relay_connection
*conn
)
3680 enum relay_connection_status status
= RELAY_CONNECTION_STATUS_OK
;
3681 struct relay_stream
*stream
;
3682 struct data_connection_state_receive_payload
*state
=
3683 &conn
->protocol
.data
.state
.receive_payload
;
3684 const size_t chunk_size
= RECV_DATA_BUFFER_SIZE
;
3685 char data_buffer
[chunk_size
];
3686 bool partial_recv
= false;
3687 bool new_stream
= false, close_requested
= false, index_flushed
= false;
3688 uint64_t left_to_receive
= state
->left_to_receive
;
3689 struct relay_session
*session
;
3691 DBG3("Receiving data for stream id %" PRIu64
" seqnum %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3692 state
->header
.stream_id
, state
->header
.net_seq_num
,
3693 state
->received
, left_to_receive
);
3695 stream
= stream_get_by_id(state
->header
.stream_id
);
3697 /* Protocol error. */
3698 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64
,
3699 state
->header
.stream_id
);
3700 status
= RELAY_CONNECTION_STATUS_ERROR
;
3704 pthread_mutex_lock(&stream
->lock
);
3705 session
= stream
->trace
->session
;
3706 if (!conn
->session
) {
3707 ret
= connection_set_session(conn
, session
);
3709 status
= RELAY_CONNECTION_STATUS_ERROR
;
3710 goto end_stream_unlock
;
3715 * The size of the "chunk" received on any iteration is bounded by:
3716 * - the data left to receive,
3717 * - the data immediately available on the socket,
3718 * - the on-stack data buffer
3720 while (left_to_receive
> 0 && !partial_recv
) {
3721 size_t recv_size
= min(left_to_receive
, chunk_size
);
3722 struct lttng_buffer_view packet_chunk
;
3724 ret
= conn
->sock
->ops
->recvmsg(conn
->sock
, data_buffer
,
3725 recv_size
, MSG_DONTWAIT
);
3727 if (errno
!= EAGAIN
&& errno
!= EWOULDBLOCK
) {
3728 PERROR("Socket %d error", conn
->sock
->fd
);
3729 status
= RELAY_CONNECTION_STATUS_ERROR
;
3731 goto end_stream_unlock
;
3732 } else if (ret
== 0) {
3733 /* No more data ready to be consumed on socket. */
3734 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64
,
3735 state
->header
.stream_id
);
3736 status
= RELAY_CONNECTION_STATUS_CLOSED
;
3738 } else if (ret
< (int) recv_size
) {
3740 * All the data available on the socket has been
3743 partial_recv
= true;
3747 packet_chunk
= lttng_buffer_view_init(data_buffer
,
3749 assert(packet_chunk
.data
);
3751 ret
= stream_write(stream
, &packet_chunk
, 0);
3753 ERR("Relay error writing data to file");
3754 status
= RELAY_CONNECTION_STATUS_ERROR
;
3755 goto end_stream_unlock
;
3758 left_to_receive
-= recv_size
;
3759 state
->received
+= recv_size
;
3760 state
->left_to_receive
= left_to_receive
;
3763 if (state
->left_to_receive
> 0) {
3765 * Did not receive all the data expected, wait for more data to
3766 * become available on the socket.
3768 DBG3("Partial receive on data connection of stream id %" PRIu64
", %" PRIu64
" bytes received, %" PRIu64
" bytes left to receive",
3769 state
->header
.stream_id
, state
->received
,
3770 state
->left_to_receive
);
3771 goto end_stream_unlock
;
3774 ret
= stream_write(stream
, NULL
, state
->header
.padding_size
);
3776 status
= RELAY_CONNECTION_STATUS_ERROR
;
3777 goto end_stream_unlock
;
3780 if (session_streams_have_index(session
)) {
3781 ret
= stream_update_index(stream
, state
->header
.net_seq_num
,
3782 state
->rotate_index
, &index_flushed
,
3783 state
->header
.data_size
+ state
->header
.padding_size
);
3785 ERR("Failed to update index: stream %" PRIu64
" net_seq_num %" PRIu64
" ret %d",
3786 stream
->stream_handle
,
3787 state
->header
.net_seq_num
, ret
);
3788 status
= RELAY_CONNECTION_STATUS_ERROR
;
3789 goto end_stream_unlock
;
3793 if (stream
->prev_data_seq
== -1ULL) {
3797 ret
= stream_complete_packet(stream
, state
->header
.data_size
+
3798 state
->header
.padding_size
, state
->header
.net_seq_num
,
3801 status
= RELAY_CONNECTION_STATUS_ERROR
;
3802 goto end_stream_unlock
;
3806 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3807 * contents of *state which are aliased (union) to the same location as
3808 * the new state. Don't use it beyond this point.
3810 connection_reset_protocol_state(conn
);
3814 close_requested
= stream
->close_requested
;
3815 pthread_mutex_unlock(&stream
->lock
);
3816 if (close_requested
&& left_to_receive
== 0) {
3817 try_stream_close(stream
);
3821 pthread_mutex_lock(&session
->lock
);
3822 uatomic_set(&session
->new_streams
, 1);
3823 pthread_mutex_unlock(&session
->lock
);
3832 * relay_process_data: Process the data received on the data socket
3834 static enum relay_connection_status
relay_process_data(
3835 struct relay_connection
*conn
)
3837 enum relay_connection_status status
;
3839 switch (conn
->protocol
.data
.state_id
) {
3840 case DATA_CONNECTION_STATE_RECEIVE_HEADER
:
3841 status
= relay_process_data_receive_header(conn
);
3843 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD
:
3844 status
= relay_process_data_receive_payload(conn
);
3847 ERR("Unexpected data connection communication state.");
3854 static void cleanup_connection_pollfd(struct lttng_poll_event
*events
, int pollfd
)
3858 (void) lttng_poll_del(events
, pollfd
);
3860 ret
= fd_tracker_close_unsuspendable_fd(the_fd_tracker
, &pollfd
, 1,
3861 fd_tracker_util_close_fd
, NULL
);
3863 ERR("Closing pollfd %d", pollfd
);
3867 static void relay_thread_close_connection(struct lttng_poll_event
*events
,
3868 int pollfd
, struct relay_connection
*conn
)
3870 const char *type_str
;
3872 switch (conn
->type
) {
3877 type_str
= "Control";
3879 case RELAY_VIEWER_COMMAND
:
3880 type_str
= "Viewer Command";
3882 case RELAY_VIEWER_NOTIFICATION
:
3883 type_str
= "Viewer Notification";
3886 type_str
= "Unknown";
3888 cleanup_connection_pollfd(events
, pollfd
);
3889 connection_put(conn
);
3890 DBG("%s connection closed with %d", type_str
, pollfd
);
3894 * This thread does the actual work
3896 static void *relay_thread_worker(void *data
)
3898 int ret
, err
= -1, last_seen_data_fd
= -1;
3900 struct lttng_poll_event events
;
3901 struct lttng_ht
*relay_connections_ht
;
3902 struct lttng_ht_iter iter
;
3903 struct relay_connection
*destroy_conn
= NULL
;
3905 DBG("[thread] Relay worker started");
3907 rcu_register_thread();
3909 health_register(health_relayd
, HEALTH_RELAYD_TYPE_WORKER
);
3911 if (testpoint(relayd_thread_worker
)) {
3912 goto error_testpoint
;
3915 health_code_update();
3917 /* table of connections indexed on socket */
3918 relay_connections_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_ULONG
);
3919 if (!relay_connections_ht
) {
3920 goto relay_connections_ht_error
;
3923 ret
= create_named_thread_poll_set(&events
, 2, "Worker thread epoll");
3925 goto error_poll_create
;
3928 ret
= lttng_poll_add(&events
, relay_conn_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
3935 int idx
= -1, i
, seen_control
= 0, last_notdel_data_fd
= -1;
3937 health_code_update();
3939 /* Infinite blocking call, waiting for transmission */
3940 DBG3("Relayd worker thread polling...");
3941 health_poll_entry();
3942 ret
= lttng_poll_wait(&events
, -1);
3946 * Restart interrupted system call.
3948 if (errno
== EINTR
) {
3957 * Process control. The control connection is
3958 * prioritized so we don't starve it with high
3959 * throughput tracing data on the data connection.
3961 for (i
= 0; i
< nb_fd
; i
++) {
3962 /* Fetch once the poll data */
3963 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
3964 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3966 health_code_update();
3968 /* Thread quit pipe has been closed. Killing thread. */
3969 ret
= check_thread_quit_pipe(pollfd
, revents
);
3975 /* Inspect the relay conn pipe for new connection */
3976 if (pollfd
== relay_conn_pipe
[0]) {
3977 if (revents
& LPOLLIN
) {
3978 struct relay_connection
*conn
;
3980 ret
= lttng_read(relay_conn_pipe
[0], &conn
, sizeof(conn
));
3984 ret
= lttng_poll_add(&events
,
3986 LPOLLIN
| LPOLLRDHUP
);
3988 ERR("Failed to add new connection file descriptor to poll set");
3991 connection_ht_add(relay_connections_ht
, conn
);
3992 DBG("Connection socket %d added", conn
->sock
->fd
);
3993 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3994 ERR("Relay connection pipe error");
3997 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4001 struct relay_connection
*ctrl_conn
;
4003 ctrl_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4004 /* If not found, there is a synchronization issue. */
4007 if (ctrl_conn
->type
== RELAY_DATA
) {
4008 if (revents
& LPOLLIN
) {
4010 * Flag the last seen data fd not deleted. It will be
4011 * used as the last seen fd if any fd gets deleted in
4014 last_notdel_data_fd
= pollfd
;
4016 goto put_ctrl_connection
;
4018 assert(ctrl_conn
->type
== RELAY_CONTROL
);
4020 if (revents
& LPOLLIN
) {
4021 enum relay_connection_status status
;
4023 status
= relay_process_control(ctrl_conn
);
4024 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4026 * On socket error flag the session as aborted to force
4027 * the cleanup of its stream otherwise it can leak
4028 * during the lifetime of the relayd.
4030 * This prevents situations in which streams can be
4031 * left opened because an index was received, the
4032 * control connection is closed, and the data
4033 * connection is closed (uncleanly) before the packet's
4036 * Since the control connection encountered an error,
4037 * it is okay to be conservative and close the
4038 * session right now as we can't rely on the protocol
4039 * being respected anymore.
4041 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4042 session_abort(ctrl_conn
->session
);
4045 /* Clear the connection on error or close. */
4046 relay_thread_close_connection(&events
,
4051 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4052 relay_thread_close_connection(&events
,
4054 if (last_seen_data_fd
== pollfd
) {
4055 last_seen_data_fd
= last_notdel_data_fd
;
4058 ERR("Unexpected poll events %u for control sock %d",
4060 connection_put(ctrl_conn
);
4063 put_ctrl_connection
:
4064 connection_put(ctrl_conn
);
4069 * The last loop handled a control request, go back to poll to make
4070 * sure we prioritise the control socket.
4076 if (last_seen_data_fd
>= 0) {
4077 for (i
= 0; i
< nb_fd
; i
++) {
4078 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4080 health_code_update();
4082 if (last_seen_data_fd
== pollfd
) {
4089 /* Process data connection. */
4090 for (i
= idx
+ 1; i
< nb_fd
; i
++) {
4091 /* Fetch the poll data. */
4092 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
4093 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4094 struct relay_connection
*data_conn
;
4096 health_code_update();
4099 /* No activity for this FD (poll implementation). */
4103 /* Skip the command pipe. It's handled in the first loop. */
4104 if (pollfd
== relay_conn_pipe
[0]) {
4108 data_conn
= connection_get_by_sock(relay_connections_ht
, pollfd
);
4110 /* Skip it. Might be removed before. */
4113 if (data_conn
->type
== RELAY_CONTROL
) {
4114 goto put_data_connection
;
4116 assert(data_conn
->type
== RELAY_DATA
);
4118 if (revents
& LPOLLIN
) {
4119 enum relay_connection_status status
;
4121 status
= relay_process_data(data_conn
);
4122 /* Connection closed or error. */
4123 if (status
!= RELAY_CONNECTION_STATUS_OK
) {
4125 * On socket error flag the session as aborted to force
4126 * the cleanup of its stream otherwise it can leak
4127 * during the lifetime of the relayd.
4129 * This prevents situations in which streams can be
4130 * left opened because an index was received, the
4131 * control connection is closed, and the data
4132 * connection is closed (uncleanly) before the packet's
4135 * Since the data connection encountered an error,
4136 * it is okay to be conservative and close the
4137 * session right now as we can't rely on the protocol
4138 * being respected anymore.
4140 if (status
== RELAY_CONNECTION_STATUS_ERROR
) {
4141 session_abort(data_conn
->session
);
4143 relay_thread_close_connection(&events
, pollfd
,
4146 * Every goto restart call sets the last seen fd where
4147 * here we don't really care since we gracefully
4148 * continue the loop after the connection is deleted.
4151 /* Keep last seen port. */
4152 last_seen_data_fd
= pollfd
;
4153 connection_put(data_conn
);
4156 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4157 relay_thread_close_connection(&events
, pollfd
,
4160 ERR("Unknown poll events %u for data sock %d",
4163 put_data_connection
:
4164 connection_put(data_conn
);
4166 last_seen_data_fd
= -1;
4169 /* Normal exit, no error */
4174 /* Cleanup remaining connection object. */
4176 cds_lfht_for_each_entry(relay_connections_ht
->ht
, &iter
.iter
,
4179 health_code_update();
4181 session_abort(destroy_conn
->session
);
4184 * No need to grab another ref, because we own
4187 relay_thread_close_connection(&events
, destroy_conn
->sock
->fd
,
4192 (void) fd_tracker_util_poll_clean(the_fd_tracker
, &events
);
4194 lttng_ht_destroy(relay_connections_ht
);
4195 relay_connections_ht_error
:
4196 /* Close relay conn pipes */
4197 (void) fd_tracker_util_pipe_close(the_fd_tracker
,
4200 DBG("Thread exited with error");
4202 DBG("Worker thread cleanup complete");
4206 ERR("Health error occurred in %s", __func__
);
4208 health_unregister(health_relayd
);
4209 rcu_unregister_thread();
4210 lttng_relay_stop_threads();
4215 * Create the relay command pipe to wake thread_manage_apps.
4216 * Closed in cleanup().
4218 static int create_relay_conn_pipe(void)
4220 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker
,
4221 "Relayd connection pipe", relay_conn_pipe
);
4224 static int stdio_open(void *data
, int *fds
)
4226 fds
[0] = fileno(stdout
);
4227 fds
[1] = fileno(stderr
);
4231 static int track_stdio(void)
4234 const char *names
[] = { "stdout", "stderr" };
4236 return fd_tracker_open_unsuspendable_fd(the_fd_tracker
, fds
,
4237 names
, 2, stdio_open
, NULL
);
4243 int main(int argc
, char **argv
)
4245 bool thread_is_rcu_registered
= false;
4246 int ret
= 0, retval
= 0;
4248 char *unlinked_file_directory_path
= NULL
, *output_path
= NULL
;
4250 /* Parse environment variables */
4251 ret
= parse_env_options();
4259 * Command line arguments overwrite environment.
4262 if (set_options(argc
, argv
)) {
4267 if (set_signal_handler()) {
4272 relayd_config_log();
4274 if (opt_print_version
) {
4280 ret
= fclose(stdin
);
4282 PERROR("Failed to close stdin");
4286 DBG("Clear command %s", opt_allow_clear
? "allowed" : "disallowed");
4288 /* Try to create directory if -o, --output is specified. */
4289 if (opt_output_path
) {
4290 if (*opt_output_path
!= '/') {
4291 ERR("Please specify an absolute path for -o, --output PATH");
4296 ret
= utils_mkdir_recursive(opt_output_path
, S_IRWXU
| S_IRWXG
,
4299 ERR("Unable to create %s", opt_output_path
);
4306 if (opt_daemon
|| opt_background
) {
4307 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
4315 if (opt_working_directory
) {
4316 ret
= utils_change_working_directory(opt_working_directory
);
4318 /* All errors are already logged. */
4323 sessiond_trace_chunk_registry
= sessiond_trace_chunk_registry_create();
4324 if (!sessiond_trace_chunk_registry
) {
4325 ERR("Failed to initialize session daemon trace chunk registry");
4331 * The RCU thread registration (and use, through the fd-tracker's
4332 * creation) is done after the daemonization to allow us to not
4333 * deal with liburcu's fork() management as the call RCU needs to
4336 rcu_register_thread();
4337 thread_is_rcu_registered
= true;
4339 output_path
= create_output_path("");
4341 ERR("Failed to get output path");
4345 ret
= asprintf(&unlinked_file_directory_path
, "%s/%s", output_path
,
4346 DEFAULT_UNLINKED_FILES_DIRECTORY
);
4349 ERR("Failed to format unlinked file directory path");
4353 the_fd_tracker
= fd_tracker_create(
4354 unlinked_file_directory_path
, lttng_opt_fd_pool_size
);
4355 free(unlinked_file_directory_path
);
4356 if (!the_fd_tracker
) {
4361 ret
= track_stdio();
4367 /* Initialize thread health monitoring */
4368 health_relayd
= health_app_create(NR_HEALTH_RELAYD_TYPES
);
4369 if (!health_relayd
) {
4370 PERROR("health_app_create error");
4375 /* Create thread quit pipe */
4376 if (init_thread_quit_pipe()) {
4381 /* Setup the thread apps communication pipe. */
4382 if (create_relay_conn_pipe()) {
4387 /* Init relay command queue. */
4388 cds_wfcq_init(&relay_conn_queue
.head
, &relay_conn_queue
.tail
);
4390 /* Initialize communication library */
4392 lttcomm_inet_init();
4394 /* tables of sessions indexed by session ID */
4395 sessions_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4401 /* tables of streams indexed by stream ID */
4402 relay_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4403 if (!relay_streams_ht
) {
4408 /* tables of streams indexed by stream ID */
4409 viewer_streams_ht
= lttng_ht_new(0, LTTNG_HT_TYPE_U64
);
4410 if (!viewer_streams_ht
) {
4415 ret
= init_health_quit_pipe();
4421 /* Create thread to manage the client socket */
4422 ret
= pthread_create(&health_thread
, default_pthread_attr(),
4423 thread_manage_health
, (void *) NULL
);
4426 PERROR("pthread_create health");
4431 /* Setup the dispatcher thread */
4432 ret
= pthread_create(&dispatcher_thread
, default_pthread_attr(),
4433 relay_thread_dispatcher
, (void *) NULL
);
4436 PERROR("pthread_create dispatcher");
4438 goto exit_dispatcher_thread
;
4441 /* Setup the worker thread */
4442 ret
= pthread_create(&worker_thread
, default_pthread_attr(),
4443 relay_thread_worker
, NULL
);
4446 PERROR("pthread_create worker");
4448 goto exit_worker_thread
;
4451 /* Setup the listener thread */
4452 ret
= pthread_create(&listener_thread
, default_pthread_attr(),
4453 relay_thread_listener
, (void *) NULL
);
4456 PERROR("pthread_create listener");
4458 goto exit_listener_thread
;
4461 ret
= relayd_live_create(live_uri
);
4463 ERR("Starting live viewer threads");
4469 * This is where we start awaiting program completion (e.g. through
4470 * signal that asks threads to teardown).
4473 ret
= relayd_live_join();
4479 ret
= pthread_join(listener_thread
, &status
);
4482 PERROR("pthread_join listener_thread");
4486 exit_listener_thread
:
4487 ret
= pthread_join(worker_thread
, &status
);
4490 PERROR("pthread_join worker_thread");
4495 ret
= pthread_join(dispatcher_thread
, &status
);
4498 PERROR("pthread_join dispatcher_thread");
4501 exit_dispatcher_thread
:
4503 ret
= pthread_join(health_thread
, &status
);
4506 PERROR("pthread_join health_thread");
4511 * Wait for all pending call_rcu work to complete before tearing
4512 * down data structures. call_rcu worker may be trying to
4513 * perform lookups in those structures.
4518 /* Ensure all prior call_rcu are done. */
4521 if (thread_is_rcu_registered
) {
4522 rcu_unregister_thread();